Interactive comment on “ SOA formation from the photooxidation of α-pinene : systematic exploration of the simulation of chamber data

The McVay et al. manuscript reports on the alpha-pinene+OH oxidation mechanism and resultant SOA, investigated using comparative measurements and modeling. Smog chamber experiments were conducted under low NO (< 2 ppb) and low and high OH (2x105 and 2x106 molecules/cm3, respectively) conditions; modeling was performed using the GECKO-A model. The GECKO-A model was updated to include gas-phase chemistry based on Vereecken et al. (2007) and dynamic gas/particle partitioning based on La et al. (2015). The significant finding was that OH levels did not influence SOA growth in the chamber studies but did influence growth in the modeling studies. Explanations for the need of a higher vapor wall loss rate in GECKO-A to match the high UV/OH experiments were explored. It was concluded that GECKO-A

is a well-studied system experimentally, for which detailed chemical mechanisms have been formulated. Here, we present the results of simulating low-NO α-pinene photooxidation experiments conducted in the Caltech chamber with the Generator for Explicit Chemistry and Kinetics of Organics in the Atmosphere (GECKO-A) under varying concentrations of seed particles and OH levels. Unexpectedly, experiments conducted at low and high OH levels yield the same SOA growth, whereas GECKO-A predicts greater SOA growth under high OH levels. SOA formation in the chamber is a result of a competition among the rates of gas-phase oxidation to low volatility products, wall deposition of these products, and condensation into the aerosol phase. Various processes, such as photolysis of condensed-phase products, particle-phase dimerization, 15 and peroxy radical autoxidation, are explored to rationalize the observations. In order to explain the observed similar SOA growth under different OH levels, we conclude that vapor wall loss in the Caltech chamber is likely of order 10 −5 s −1 , consistent with previous experimental measurements in that chamber. We find that GECKO-A tends to overpredict the contribution to SOA of later-generation oxidation products under high
Fewer studies have examined SOA formation from OH oxidation of α-pinene, and even 10 fewer studies have focused on low-NO conditions, under which most α-pinene is oxidized (Nozière et al., 1999;Ng et al., 2007a;Claeys et al., 2009;Eddingsaas et al., 2012a). However, nearly half of the global sink of α-pinene has been estimated to be reaction with OH (42 %); the remainder is oxidized predominantly by O 3 , with only 12 % oxidized by NO 3 (Capouet et al., 2008). 15 As in virtually all volatile organic compound (VOC) systems that lead to SOA, the OH oxidation of α-pinene is complex, and parts of the mechanism are still not fully characterized (Peeters et al., 2001;Vereecken et al., 2007;Eddingsaas et al., 2012a, b). Theoretical and experimental work has indicated that heretofore unidentified alkoxy and peroxy isomerization channels may contribute substantially to SOA formation (Peeters et several non-traditional pathways in the OH oxidation of α-pinene based on quantumand theoretical chemical calculations. Species in Fig. 1 are labeled to be consistent with Vereecken et al. (2007). Species R1 is formed with 22 % yield by a prompt ring-opening of the four-membered ring following initial OH addition to the double bond in α-pinene (Peeters et al., 2001). A 40 : 60 ratio of syn vs. anti stereochemistry is estimated for 10 species R1 (Vereecken et al., 2007). The anti conformer is predicted to undergo a 1,6hydrogen shift to form a peroxide, forming a radical stabilized by allyl resonance, R7. This reaction is predicted to dominate over all other peroxy reactions for the anti conformer even in high-NO conditions. The position of oxygen addition to the alkyl radical R7 is not well-constrained. The branching ratio between addition at the secondary and 15 tertiary sites has conservatively been predicted to be 50 : 50 (Vereecken et al., 2007). Vereecken et al. (2007) calculate that the syn conformer of R1 will participate in traditional peroxy chemistry but will also undergo ring closure to form a six-membered ring, species R3. Vereecken et al. (2007) calculate a rate constant for this reaction of 2.6 s −1 at 298 K. Species R2 in Fig. 1, formed from the reaction of the syn conformer of 20 species R1 with NO and RO 2 , is predicted to undergo a ring-closure reaction to form species R10. This reaction is predicted to dominate over acetone elimination and hydroxy formation, the previous pathways for R2 in GECKO-A (Valorso et al., 2011). R10 adds oxygen and then participates in the usual peroxy reactions. The HO 2 chemistry has also been updated in GECKO-A. The main reaction channel 25 for reaction of alkyl peroxy radicals with HO 2 is formation of hydroperoxides. Recent evidence indicates that the reaction of some peroxy radicals with HO 2 can also lead to an alkoxy radical, regenerating OH (Hasson et al., 2004;Dillon and Crowley, 2008;Birdsall et al., 2010;Eddingsaas et al., 2012a). This reaction is promoted by the presence of 33165 neighboring polar functional groups, which can stabilize the intermediate leading to the alkoxy through hydrogen bonding (Hasson et al., 2005). This pathway has been added to GECKO-A for molecules with an oxygenated moiety in the α position to the peroxy radical. Estimates of the yield of the alkoxy for small acyl peroxy radicals range from 0. 15-0.7 (Hasson et al., 2004;Dillon and Crowley, 2008); however, branching ratios 5 for the larger molecules in the α-pinene mechanism have not been determined. In light of this uncertainty, the branching ratio between the hydroperoxide and the alkoxy has been assigned to 80 : 20 in GECKO-A. The above updates constitute the base mechanism. Sensitivity tests applied on this base mechanism are addressed in the sections that follow.

Condensed-phase photolysis
Photolysis of gas-phase compounds containing carbonyl, peroxide or nitrate chromophores is included automatically within oxidation schemes generated by GECKO-A (Aumont et al., 2005). For the simulations presented here, compound-specific photolysis rates are calculated using cross sections and quantum yields described in Au-15 mont et al. (2005) and measured irradiance data in the Caltech chamber. Photolysis of condensed-phase compounds, recently shown to lead to rapid loss of α-pinene SOA (Henry and Donahue, 2012;Donahue et al., 2012;Epstein et al., 2014;Wong et al., 2015), is included here as a possible reaction route. Because GECKO-A does not generate chemical reactions within the condensed phase, radical species pro-20 duced by condensed-phase photolytic reactions are assumed to be irreversibly lost and to not participate in subsequent chemistry . This approximation likely represents an upper limit to the actual physical process since it does not account for recombination of fragments in the particle phase (Henry and Donahue, 2012;Hodzic et al., 2015). Two methods of calculating condensed-phase photolysis 25 rate constants were tested in Hodzic et al. (2015): (1) all photolabile compounds in the condensed phase are assigned the same empirically-derived rate constant, and (2) the condensed-phase photolytic rate constant for each species is set to the corresponding 33166 gas-phase value. Hodzic et al. (2015) showed that the second method generally results in a greater loss of SOA. Therefore, assumption (2) is tested in the present simulations to evaluate the maximum possible impact of condensed-phase photolysis on overall SOA growth.

5
The importance of aerosol-phase dimerization and oligomerization reactions has been demonstrated (Kalberer et al., 2004;Gao et al., 2004a, b;Tolocka et al., 2004;Yee et al., 2013;Shiraiwa et al., 2013;Fahnestock et al., 2014). These include alcohol + carbonyl to form hemiacetals and acetals, hydroperoxide + carbonyl to form peroxyhemiacetals and peroxyacetals, carboxylic acid + alcohol to form esters, and aldehyde 10 self-reactions to form aldols (Ziemann and Atkinson, 2012). Shiraiwa et al. (2013) modeled a generalized particle-phase reaction of an SVOC + carbonyl with a rate constant of 12 M −1 s −1 when fitting the observed evolution of the particle size distribution of dodecane SOA. In contrast, Capouet et al. (2008) modeled the formation of peroxyhemiacetals during α-pinene SOA formation using a rate constant of 0.06 M −1 s −1 ; they 15 found that these reactions had a minor impact on SOA yield, except during ozonolysis with large VOC/NO x ratios that led to high hydroperoxide yields. In the present work, particle-phase dimerization reactions have been added to the GECKO-A α-pinene oxidation scheme for aldehydes, hydroperoxides, alcohols, and carboxylic acids. Three reaction pairings are considered: aldehyde + alcohol, aldehyde 20 + hydroperoxide, and carboxylic acid + alcohol. As a first approximation, reactions are assumed to be irreversible and to form nonvolatile products. The mass of dimers becomes part of the organic aerosol mass into which gas-phase species can partition. The rate constant is set to be the same for each dimerization reaction and two values are tested: 12 M −1 s −1 (Shiraiwa et al., 2013) and 0.01 M −1 s −1 (Ziemann and Atkinson, In previous versions of GECKO-A, instantaneous equilibrium partitioning is assumed to occur between the vapor and particle phases (Camredon et al., 2007;. As a more general treatment of vapor-particle transport, dynamic partitioning for mass transfer between the gas and the particle phase is implemented here as described in La et al. (2015), with a few modifications. Mass transfer to and from the 10 particle is represented by where k gp and k pg are first-order rate constants for transport to and from the particle (s −1 ), and G i and P i are concentrations in the gas and particle phase, respectively.

15
Mass transfer can be limited by either gas-phase diffusion, interfacial accommodation, or particle-phase diffusion (e.g., Mai et al., 2015). Particle-phase diffusion in GECKO-A is assumed to be sufficiently rapid that particles are well-mixed. Recent studies have shown that the assumption of rapid particle-phase diffusion in SOA may not be accurate owing to semi-solid behavior (Perraud et al., 2012;Cappa and Wilson, 2011;20 Vaden et al., 201020 Vaden et al., , 2013Virtanen et al., 2010Virtanen et al., , 2011Kuwata and Martin, 2012;Saukko et al., 2012;Abramson et al., 2013;Renbaum-Wolff et al., 2013). Computationally, the gas-particle accommodation coefficient, α p , is used to approximate resistances to gas-33168 Full Screen / Esc

Printer-friendly Version
Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | particle partitioning from surface accommodation and particle-phase diffusion (Saleh et al., 2013). Therefore, the overall rate constant of mass transfer k gp can be approximated as with k diff = 4πD g r p C p and k int = α pc πr 2 p C p , where D g is the species gas-phase diffu-5 sivity, r p is the particle radius, C p is the number of particles per unit volume of air, α p is the gas-particle accommodation coefficient, andc is the gas-phase mean velocity (Seinfeld and Pandis, 2006). In the present work, the accommodation coefficient α p is treated as a parameter that can be varied in order to produce the best fit to the observed SOA growth. In the model, the number concentration of particles C p remains 10 unchanged over the course of the experiment as r p grows owing to condensation of organic aerosol. Particle wall loss is not included in the model. SOA growth in the α-pinene system is essentially independent of seed surface area (discussed in the Results section); therefore, lack of particle wall loss in GECKO-A will not substantially affect SOA predictions. GECKO-A does not currently include a mechanism for nucle-15 ation; therefore, for the nucleation experiment, an initial seed concentration of 10 4 cm −3 was assumed with an initial radius of 5 nm. r p is calculated at each time step as where r p0 is in the initial inorganic seed radius, M aer is the organic aerosol mean molecular weight, C aer is the organic aerosol mass concentration (molec cm −3 ), and ρ OA is 20 the organic aerosol density (1.32 g cm −3 for α-pinene SOA (Ng et al., 2007a)). Condensation to and evaporation from the particle occurs until equilibrium is reached between the gas and particle phases (if ever). The gas-particle partitioning at equilibrium is assumed to follow Raoult's law; therefore the reverse rate constant for mass transfer from 33169 Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the particle to the gas-phase is calculated from the relationship (La et al., 2015) k gp where R is the ideal gas constant, T is the temperature, P vap is the species vapor pressure, and γ aer is the activity coefficient in the particle phase (assumed to be 1).

5
Chamber data are affected by wall deposition of semi and low-volatility vapors (Loza et al., 2010;Matsunaga and Ziemann, 2010;Kokkola et al., 2014;Ziemann, 2014, 2015;Zhang et al., 2014Zhang et al., , 2015Bian et al., 2015). Partitioning of vapor species between the gas phase and the chamber wall is based on the parameterization of Matsunaga and Ziemann (2010) and is implemented as described in La et al. (2015), with minor variations. Reported values for the rate of transport from the gas to the wall, k gw , span several orders of magnitude in different chambers (Matsunaga and Ziemann, 2010;Zhang et al., 2014Zhang et al., , 2015Loza et al., 2010Loza et al., , 2014. Wall loss rates directly measured in the Caltech chamber are of order 10 −5 -10 −6 s −1 (Zhang et al., 2015;Loza et al., 2010Loza et al., , 2014, although Zhang et al. (2014) fit a wall loss rate of 10 −4 s −1 during 15 toluene SOA experiments. Gas-wall equilibration timescales, τ g,w , of order 10-100 min have consistently been measured in the 8.2 m 3 chamber used by Yeh and Ziemann (2015). With τ g,w = 1/(k gw + k wg ) (La et al., 2015), these timescales correspond to wall loss rates of 10 −3 -10 −4 s −1 . The Yeh and Ziemann (2015) chamber is smaller than the Caltech chamber, 8.2 vs. 24 m 3 ; however, the differences in measured wall loss rates 20 are greater than expected if wall loss scales with the surface-to-volume ratio of the chamber. It has been suggested that rapid initial vapor wall loss could be difficult to distinguish from changes in concentration due to injection and mixing during the long fill period in the Caltech chamber Loza et al., 2014). Vapor species could potentially rapidly equilibrate with the wall and be subsequently lost to the wall with a much slower uptake owing to relaxation of the Teflon polymer (Yeh and Ziemann, 2015). Sensitivity tests in the present work examine the extent to which these differences indicate significantly slower vapor wall loss in the Caltech chamber or arise from the procedure by which experimental measurements were carried out. Furthermore, Zhang et al. (2015) showed that k gw may depend on the species' volatility; here we 5 assume k gw to be the same for all species in order to evaluate the order of magnitude of k gw needed to fit the observations. The nominal value is set to 10 −3 s −1 . As an additional sensitivity test, we implement the k gw parameterization developed by Zhang et al. (2015) in which k gw increases as species' volatility decreases. Transport between the gas-phase and the wall is represented by the balance equations, where k gw and k wg are first-order rate constants for transport to and from the wall, and W i is the concentration in the wall layer. The reverse rate is calculated assuming gas-wall partitioning follows Raoult's law (La et al., 2015): where C w is the equivalent overall organic mass concentration in the wall, M w is the equivalent molar weight of the organic concentration in the wall, and γ w is the activity coefficient in the Teflon film. Values of C w /M w γ w must be determined experimentally from chamber observations (Matsunaga and Ziemann, 2010;Zhang et al., 2015). In the 20 present simulations, a nominal C w /M w γ w value of 120 µmol m −3 , the value determined by Matsunaga and Ziemann (2010) for 2-ketones, is used for all species except for α-pinene, for which C w /M w γ w = 20 µmol m −3 , the value determined by Matsunaga and Ziemann (2010) for alkenes. C w /M w γ w is varied as a sensitivity test. 33171

Experimental
Six α-pinene photooxidation experiments were conducted in the Caltech dual 24 m 3 environmental chambers at ∼ 298 K and < 5 % RH (Table 1). Prior to each experiment, the Teflon chambers were flushed with purified, dry air for 24 h until the particle number concentration was < 10 cm −3 and volume concentration was < 0.01 µm 3 cm −3 . Hydro-5 gen peroxide (H 2 O 2 ) was used as the OH source by evaporating 113 µL of 50 wt % aqueous solution into the chamber with 5 L min −1 purified air for 110 min, resulting in an approximate starting H 2 O 2 mixing ratio of 2 ppm. Experiments were conducted at low NO (< 2 ppb). Heated 5 L min −1 of purified, dry air flowed through the glass bulb into the chamber for 30 min, introducing ∼ 50 ppb α-pinene into the chamber. Ammonium sulfate (AS) seed aerosol was injected into the chamber by atomizing 0.015 or 0.1 M aqueous (NH 4 ) 2 SO 4 solution into the chamber for 30 to 90 min, varying the initial AS concentration in order to vary the seed surface area (SA) available for condensation. After ∼ 1 h mixing, photooxidation was initiated. Four of the experiments were performed using the full set of blacklights, for which the calculated j NO 2 = 4×10 −3 s −1 , with 15 varying amounts of inorganic seed particles (Table 1). The remaining two experiments were performed using only 10 % of the available blacklights, for which the calculated j NO 2 = 3.7 × 10 −4 s −1 , again with varying amounts of inorganic seed particles.
Relative Humidity (RH) and temperature were monitored via a Vaisala HMM211 probe. O 3 and NO x mixing ratios were measured by a Horiba O 3 analyzer (APOA-360) 20 and a Teledyne NO x analyzer (T200), respectively. α-pinene concentration was monitored by a gas chromatograph equipped with a HP-5 column (15 m × 0.53 mm ID × 1.5 µ m thickness, Hewlett-Packard) coupled with flame ionization detector (GC/FID, Agilent 6890N). The size distribution and number concentration of seed particles and organic aerosols were characterized using a custom-built Scanning Mobility Particle  Volume distributions are corrected for particle wall loss using size-dependent wall loss rates measured in the Caltech chamber. Two limiting assumptions have traditionally been made when correcting for particle wall loss (Loza et al., 2012): In the lower bound, once deposited, particles are assumed to no longer interact with the vapor. The mass of SOA present on a particle at the moment of its deposition is added when 5 calculating the total SOA. In the upper bound, deposited particles are assumed to continue growing at the same rate as suspended particles, and this SOA is similarly added to the total SOA. The upper bound can be viewed as an early approximation of vapor wall loss from a time when vapor wall loss was less well understood. However, this approximation does not account for the differing mechanisms of vapor-particle and vapor-wall transport, which is reflected in different time scales. Moreover, the absorbing mass of the wall, represented by the parameter C w , is ∼ 3 orders of magnitude higher than the mass of deposited particles . The upper bound is therefore less appropriate to use when vapor wall loss is accounted for separately, and thus the lower bound is used here. The upper bound is shown for reference in Fig. S3 in the 15

Supplement.
The initial concentrations of H 2 O 2 , NO x , and O 3 in the chemical model were optimized to match the α-pinene decay and O 3 formation. The initial mixing ratio of H 2 O 2 was optimized to be 2 ppm, yielding predicted OH levels of ∼ 2 × 10 6 molec cm −3 for the 100 % UV experiments and ∼ 2 × 10 5 molec cm −3 for the 10 % UV experiments. Al-20 though experiments were conducted at low-NO conditions, background levels of NO x were observed in the chamber. Due to voltage interference in the NO x measurements at such low values, it was not possible to determine the exact amount of NO x present in the chamber. Therefore, the background level in the chemical model was set to 0.1 ppb, split evently between NO and NO 2 . Ozone formation was observed during the 25 experiments, giving evidence of the presence of NO 2 in the chamber. Because any NO 2 initially present in the chamber is quickly photolyzed, a NO x wall off-gassing rate of 2.5 ppt min −1 is needed to match the observed ozone formation (e.g., Valorso et al., 2011). Predicted SOA formation is not sensitive to the assumed NO x off-gassing rate.

Results
Observed SOA growth curves as a function of reacted α-pinene are shown in Fig. 2 for the six photooxidation experiments. Corresponding temporal plots of SOA are shown in Fig. S3. (Gaps in SOA data resulted when the DMA was taken off line briefly to clean the inlet.) With the exception of the nucleation experiment, SOA growth is observed immediately upon irradiation, similar to Ng et al. (2007a). When presented as a function of reacted α-pinene, the growth curves essentially overlap, regardless of the UV level or the initial seed particle concentration. Overlap at two different UV intensities indicates that SOA growth is not very sensitive to the oxidation rate for these conditions. SOA yields (ratio of maximum mass of SOA formed to mass of α-pinene reacted) for these 15 experiments range from 17 to 26 %, at the low end of the range of previously reported yields of 26-45 % for this system (Ng et al., 2007a;Eddingsaas et al., 2012b). Figure 2 also shows predictions from the base GECKO-A mechanism for the six experiments. In the GECKO-A model, the two main parameters representing vaporparticle and vapor-wall transport, α p and k gw , are set to nominal values of 1 and 20 10 −3 s −1 , respectively. C w /(M w γ w ) is set to 120 µmol m −3 for all species except αpinene. These parameters give a good agreement for the 100 % UV experiments. Figure S4 in the Supplement shows the effect of varying the vapor-particle accommodation coefficient α p for different values of the vapor wall loss rate k gw . α p = 1 or 0.1 yield almost identical SOA predictions for all experiments except nucleation. In the nucleation case, α p = 0.1 results in substantially less SOA predicted, which may result because nucleation is not treated explicitly in GECKO-A but instead approximated by 33174 Introduction initializing the particle concentration with particles of 5 nm radius. Lowering the value of α p to 0.01 or 0.001 delays the onset of SOA formation for the 100 % UV experiments, which is not consistent with experimental observations. Lowering k gw results in substantial overprediction of SOA for the 100 % UV experiments. For the 10 % UV experiments, all combinations of parameters underpredict the SOA. Figure S5 in the 5 Supplement shows the effect of varying C w /(M w γ w ) for different values of the vapor wall loss rate k gw . C w /(M w γ w ) controls partitioning between the gas phase and the wall at equilibrium; therefore, variations in C w /(M w γ w ) have more of an effect on SOA predictions when using a faster wall loss rate because equilibrium is approached sooner. The base C w /(M w γ w ) and k gw = 10 −3 s −1 give the best agreement for 100 % UV experiments; however, SOA for the 10 % UV experiments is still underpredicted. This discrepancy will be addressed subsequently. All subsequent simulations are conducted with α p = 1 and the base C w /(M w γ w ).
The best-fit α p = 1 (and 0.1 for all experiments except nucleation) for the 100 % UV experiments suggests that there are no substantial limitations to vapor-particle mass 15 transfer (Mai et al., 2015). This conclusion is consistent with the experimental observation that the SOA growth in this system is virtually independent of the amount of seed surface area present; dependence of SOA growth on seed surface area occurs only with diffusion-or accommodation-limited vapor-particle mass transfer . Optimal α p = 1 or 0.1 is consistent with Saleh et al. (2013) who determined 20 accommodation coefficients of order 0.1 for α-pinene ozonolysis SOA. This result differs from that of Zhang et al. (2014), who observed toluene SOA formation to depend strongly on the seed surface area and consequently to adhere to a low α p = 0.001. The different behavior observed for α-pinene and toluene SOA may reflect differences in SOA formation mechanisms; however, it is not possible to discern the reason for the 25 difference based solely on the data at hand. If α-pinene SOA behaves as a semi-solid (Perraud et al., 2012;Cappa and Wilson, 2011;Vaden et al., 2010Vaden et al., , 2013Virtanen et al., 2010Virtanen et al., , 2011Kuwata and Martin, 2012;Saukko et al., 2012;Abramson et al., 2013;Renbaum-Wolff et al., 2013) formed, SOA is converted into a glassy state through hydrogen bonding or oligomerization (Cappa and Wilson, 2011), SOA formation itself could still be characterized by a high α p . The observation that SOA growth is insensitive to oxidation rate (shown by the overlap of the growth curves in Fig. 2 for different UV exposures) is not consistent with 5 predictions regarding the interplay of reaction rate and vapor wall loss: if the reaction rate increases relative to the vapor wall loss rate, successive generations of low volatility species will be produced more quickly and will condense preferntially onto particles before their precursors are lost to the chamber walls, leading to a higher SOA yield McVay et al., 2014). Such behavior was observed experimentally by Ng et al. (2007b): higher SOA yields were observed in the aromatic system using HONO as an OH precursor than during those in which only NO and NO 2 are present initially, leading to much lower OH levels. This "rate effect" was tentatively attributed by Ng et al. (2007b) to loss of semivolatile organics to the chamber walls, although at a time when the nature of vapor wall deposition was less well understood. A dependence 15 of SOA growth on oxidation rate is also predicted by the base scenario in GECKO-A: the SOA levels are predicted to be substantially lower at 10 % UV owing to significant organic mass loss to the walls. A number of explanations can be advanced for the overlap of the growth curves at high and low UV. These possibilities are illustrated schematically in Fig. 3 and are systematically explored below.

Negligible or slow vapor wall loss
In the absence of vapor wall loss, the growth curves predicted by GECKO-A overlap at high and low UV (Fig. S6 in the Supplement). Oxidation occurs more slowly under low UV owing to the decreased rate of generation of OH, but if condensable species or their precursors do not deposit on the walls, the same amount of SOA eventually forms.
However, in the absence of vapor wall loss, GECKO-A overpredicts SOA substantially for the high UV experiments and slightly for the low UV experiments. At a low level of vapor wall loss, k gw = 10 −5 s −1 , the growth curves predicted by GECKO-A still essentially overlap (Fig. 4). A "rate effect", in which the SOA yield depends on the rate of oxidation, is observed only if the rate of oxidation is slower than or competitive with other loss processes. The rate constant for the reaction of OH with α-pinene is 5.23 × 10 −11 cm 3 molec −1 s −1 (Atkinson and Arey, 2003). OH concentra-5 tions predicted by GECKO-A of 2 × 10 6 and 2 × 10 5 molec cm −3 for the high and low UV experiments, respectively, give overall reaction rates of 10 −4 and 10 −5 s −1 . (OH reaction rate constants for oxidized products will be slower than that for α-pinene, and the reaction rate slows in later generations). A vapor wall loss characterized by k gw = 10 −3 s −1 exceeds substantially either of these reaction rates, and thus at this 10 wall deposition rate the effect of changing oxidation rate is strong. At a much slower wall loss, k gw = 10 −5 s −1 , the rate effect is less pronounced, leading to overlap of the growth curves predicted by GECKO-A. The absence of a rate effect in the experimental observations is consistent with a slower vapor wall loss rate, in accord with wall loss rates that have been measured previously in the Caltech chamber (Loza et al., 2010, 15 2014; Zhang et al., 2015). With a vapor wall loss rate of 10 −5 s −1 , SOA predictions match the data fairly well at low UV, within 10 µg m −3 of the final SOA concentration, but remain overpredicted by 80 to 200 µg m −3 at high UV. Thus, a slow wall loss rate alone is not sufficient to reconcile the predictions at high and low UV. However, it is instructive to examine the 20 contribution of different generations of reaction to SOA predictions with this slow vapor wall loss (Fig. 5). GECKO-A predicts that SOA will consist of almost entirely secondgeneration products. At the end of the experiment, simulations under high UV predict ∼100 µg m −3 of second-generation products, as opposed to ∼50 µg m −3 for low UV. Thus, reducing the contribution of second-generation products in high UV may result 25 in closer model-data agreement for both high and low UV conditions at a slow vapor wall loss rate.

Overcontribution of second and later generation species
As a sensitivity test, all OH reaction rate constants, except that of OH + α-pinene, were reduced by varying factors. Figure S7 in the Supplement shows the impact of reducing the OH rate constants to 1, 10, and 50 % of the default values at a vapor wall loss rate of 10 −5 s −1 . A reduction to 10 % and a wall loss rate of 10 −5 s −1 results in the best fit 5 to both high and low UV experiments (Fig. 6), albeit with a slight overprediction for the high UV experiments and slight underprediction for the low UV experiments. While this result does not necessarily suggest that reaction rate constants in GECKO-A are overpredicted by an order of magnitude, it does suggest that second and higher-generation compounds may be primarily responsible for the model-measurement discrepancy at differing UV levels. An excess contribution of later-generation compounds to SOA in GECKO-A could be due to several factors: (1) overprediction of the OH reaction rate constants by the SARs, (2) assumed reaction pathways in GECKO-A, (3) underestimation of volatilities for later-generation species, or (4) significantly faster wall loss rate for later generation products. Each of these possibilities is considered in turn. 15 OH reaction rate constants in GECKO-A are based on SARs from Kwok and Atkinson (1995) and subsequent updates, which have been shown to predict OH reaction rate constants within a factor of 2 for alkanes, alkenes, and diols, but may be more uncertain for the complex and highly functionalized compounds in later generations of α-pinene oxidation (Bethel et al., 2001). For example, Bethel et al. (2001) showed that 20 OH rate constants for 1,2-hydroxyaldehydes are overestimated by factors of 3-4 using the established SARs. Although rate constants are likely not overpredicted by an order of magnitude as the simulations suggest, it is possible that uncertainties in the SARs for later-generation species may contribute to the overprediction of these products by GECKO-A. 25 SARs used to predict reaction pathways and branching ratios in GECKO-A are also prone to uncertainties. In the reaction of peroxy radicals with HO 2 , the dominant pathway in GECKO-A is formation of the hydroperoxide (assumed branching ratio 80 %). α-33178 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | pinene hydroxy dihydroperoxide (C 10 H 16 O 5 ) is a second-generation compound formed via two such reactions with HO 2 that is predicted by GECKO-A to contribute significantly to SOA at the high UV conditions. Figure S8 in the Supplement shows that eliminating the mass of this compound from SOA predictions in the absence of vapor wall loss results in a much more significant decrease in SOA at high UV (∼ 100 µg m −3 ) 5 than at low UV (∼ 15 µg m −3 ). α-pinene hydroxy dihydroperoxide results from two successive OH reactions and is therefore formed more rapidly and in greater amounts at a higher OH level. α-pinene hydroxy dihydroperoxide has been observed in the gas phase with an estimated 3 % yield (Eddingsaas et al., 2012a). In the high UV simulation, this compound is formed with >30 % yield; therefore, it is likely that reaction 10 pathways leading to this product are overrepresented in GECKO-A.
A second pathway for the reaction of peroxy radicals with HO 2 is formation of an alkoxy and regeneration of OH, added to GECKO-A with an uncertain branching ratio of 20 % and only for peroxy radicals with an oxygenated moiety in the α position. Simulations show that SOA predictions are not sensitive to the second assumption but 15 are very sensitive to the assumed branching ratio for this reaction (Fig. 7). Increasing formation of the alkoxy leads to decreasing SOA formation, with a greater effect at high UV than at low UV. Uncertainty in this branching ratio in the α-pinene system could be a major factor leading to the high/low UV discrepancy.
Finally, Praske et al. (2015) recently identified a third HO 2 reaction channel for per-20 oxy radicals produced from methyl vinyl ketone (MVK): formation of a carbonyl and regeneration of OH and HO 2 . In place of a hydroperoxide, which results from the standard HO 2 reaction channel, this reaction produces a ketone, which has a higher vapor pressure and therefore forms less SOA. If this reaction occurs in the α-pinene system, the absence of this pathway in GECKO-A could contribute to SOA overprediction. exist for vapor pressures of semi-and low-volatility species, and uncertainty in vapor pressure estimation increases as vapor pressure decreases (Valorso et al., 2011). Therefore, underprediction of vapor pressures for later-generation species could lead to excess SOA. Indeed, Valorso et al. (2011) showed that significant uncertainties in SOA predictions result when using different methods for vapor pressure estimation; 5 however, the Nannoolal et al. (2008) method generally led to the highest volatilites and the lowest SOA predictions when compared to the Myrdal and Yalkowsky (1997) method and the SIMPOL-1 method from Pankow and Asher (2008). Therefore, underprediction of vapor pressure is likely not responsible for the enhanced contribution of later-generation species to SOA.

10
A final possible explanation for the overprediction of higher-generation SOA species is that significantly faster vapor wall loss exists for these more functionalized species. Zhang et al. (2015) showed that vapor wall loss rates increase as compound vapor pressure decreases. They measured wall loss rates for primarily first-generation oxidation products to be 10 −5 -10 −6 s −1 in the Caltech chambers. Implementing the k gw 15 parameterization developed by Zhang et al. (2015) in GECKO-A yields similar SOA predictions as those based on assuming a fixed k gw = 10 −5 s −1 (with the exception of the nucleation experiment, in which the k gw parameterization predicted more SOA than using a fixed k gw = 10 −5 s −1 ). However, it is certainly possible that later generation, more functionalized compounds could exhibit significantly faster wall loss than predicted by 20 the parameterization of Zhang et al. (2015). As one example, Krechmer et al. (2015) fit a wall loss rate of 3 × 10 −3 s −1 for a C 5 H 10 O 5 isoprene oxidation product in the Caltech chambers. If later-generation species in the α-pinene system exhibit markedly higher wall loss as well, SOA overprediction in GECKO-A would be reduced. In summary, if GECKO-A overpredicts the contribution of later-generation species, 25 the measurement-model discrepancy at differing UV levels could be reconciled. Reducing the contribution of second-generation species also requires a slower vapor wall loss rate to fit the data, which is actually more consistent with those that have been measured in the Caltech chamber.

Condensed-phase photolysis
If condensed-phase photolysis occurs in the chamber but is missing in GECKO-A, both the high/low UV measurement-model discrepancy and the high wall loss rate needed to match the 100 % UV observations could potentially be explained. If condensedphase photolysis is an efficient loss mechanism for SOA at 100 % UV, a lower vapor 5 wall loss rate would be required in order to continue to fit experimental observations. Because condensed-phase photolysis would be less efficient at 10 % UV, under this slower wall loss rate the low UV SOA predictions will increase in the direction of the observations. However, this effect is not seen in the simulations. Figure 8 shows the effect of condensed-phase photolysis, with the radicals produced assumed to be lost 10 permanently. Two wall loss rates are shown, the default k gw = 10 −3 s −1 and a slower k gw = 10 −4 s −1 . Only one high UV experiment is shown, but the results are similar for all high UV experiments. Moreover, condensed-phase photolysis has no effect on the low UV predictions (not shown). At high UV conditions, the assumed presence of condensed-phase photolysis leads 15 to a slight reduction in SOA predictions at the end of the experiment, regardless of k gw . Note that the current implementation of condensed-phase photolysis represents an upper limit: compounds in the condensed phase are taken to photolyze at the same rate as in the gas phase, and fragmentation products are lost permanently. The minor effect of this photolysis upper limit can be explained by comparing the wavelength-20 dependent photon flux of the chamber to absorption cross sections of species predicted to constitute the SOA (Fig. S9 in the Supplement). Wong et al. (2015) found that the main absorption region of SOA produced by α-pinene ozonolysis corresponds to characteristic absorptions of carbonyl and peroxide functional groups and that the SOA absorption is strongest between 240 and 400 nm. Although products of α-pinene 25 OH oxidation and ozonolysis differ, peroxide and carbonyl functional groups are generated during both (Eddingsaas et al., 2012a). The photon flux in the Caltech chamber is less intense than the solar flux in this region, particularly for the wavelength re- gion over which peroxides photolyze (Fig. S9). Because peroxides are expected to be abundant in α-pinene SOA (Capouet et al., 2008;Eddingsaas et al., 2012a), the effect of condensed-phase photolysis may be underrepresented in the Caltech chamber as compared to the atmosphere. The minor effect of condensed-phase photolysis in these simulations is consistent 5 with the findings of Hodzic et al. (2015), that reduction in SOA from both gas-phase and condensed-phase photolysis for various systems, including α-pinene, is minor during the initial 10 h of a simulation but becomes substantial over a week of atmospheric ageing. Moreover, this result indicates that this process does not explain the high/low UV measurement-model discrepancy. Even in the presence of condensed-phase pho-10 tolysis, at a wall loss rate k gw = 10 −4 s −1 a strong overprediction of SOA still exists at 100 % UV. Although condensed-phase photolysis is potentially important in the atmosphere on longer timescales , it does not explain the low UV measurement-model discrepancy of these experiments. 15 Autoxidation has been demonstrated to be important in SOA formation (Crounse et al., 2013;Ehn et al., 2014;Rissanen et al., 2014;Jokinen et al., 2014;Mentel et al., 2015;Rissanen et al., 2015). Formation of extremely low volatile organic compounds (ELVOCs) via autoxidation has been observed from the OH oxidation of α-pinene, although in a lower yield than from ozonolysis (Ehn et al., 2014;Jokinen et al., 2014).

20
In autoxidation, a peroxy radical undergoes an intramolecular hydrogen abstraction to form a hydroperoxide, generating an alkyl radical which then adds oxygen to reform a peroxy radical. This pathway becomes important only when the rate of hydrogen abstraction is competitive with bimolecular reactions of the peroxy radicals with NO, HO 2 , and RO 2 (Crounse et al., 2013). Because lower UV intensities will lead to lower HO 2 25 and RO 2 concentrations, intramolecular hydrogen abstraction could potentially be favored under low UV. For the high UV experiments, the predicted lifetime with HO 2 is ∼10 s; for the low UV experiments, ∼50 s, indicating autoxidation may potentially be 33182 concentration and will not be produced slower under low UV. Therefore, adding autoxidation pathways will likely increase the fraction of SOA composed of first-generation products, which may lead to similar predictions for high and low UV. However, explicit mechanisms and rate constants are still lacking for autoxidation in the α-pinene system, although they have been developed for simpler cycloalkenes (Mentel et al., 2015;Rissanen et al., 2015). In the absence of these explicit pathways, this hypothesis cannot be currently tested. However, autoxidation remains a likely process to explain the overlap of the growth curves at high and low UV.

Particle-phase dimerization
If dimerization is more competitive with gas-phase fragmentation at the lower OH lev-15 els at 10 % lights, increased SOA growth could result. Figure 9 shows one 100 % UV and one 10 % UV experiment with two different values for the dimerization rate constant: 0.01 M −1 s −1 (Ziemann and Atkinson, 2012;Capouet et al., 2008) and 12 M −1 s −1 (Shiraiwa et al., 2013) using the same base k gw = 10 −3 s −1 . The 12 M −1 s −1 rate constant results in significant SOA overprediction for both the 100 % and the 10 % UV 20 experiments; the slower dimerization rate constant still results in an overprediction for the 100 % UV experiment but an underprediction at 10 % UV. For this rate constant, rapid vapor wall loss combined with the slower chemistry at low UV prevents significant SOA formation, despite dimerization. In the high UV experiment, because gasphase chemistry occurs faster relative to vapor wall loss, even a slower dimerization 25 rate increases SOA predictions. While dimerization increases SOA predictions for low UV, it has the same effect at high UV. Therefore, dimerization in its current implemen-33183 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | tation in GECKO-A does not resolve the low UV discrepancy without causing an SOA overprediction for high UV.

Enhanced wall loss at high UV
To recapitulate, using the base chemistry in GECKO-A, predictions agree with the high UV experiments when using a relatively fast vapor wall loss rate of 10 −3 s −1 , and predic-5 tions approximately match the low UV experiments when using a much slower wall loss rate of 10 −5 s −1 . This result prompts the hypothesis that vapor wall loss could somehow be dependent on the UV level. The diffusion coefficient and the eddy diffusivity could change based solely on the UV level if a change in the UV level results in a temperature change in the chamber. The temperature of the chamber was ∼ 3 K higher for the 10 high UV experiments than for the low. Gas-phase diffusion coefficients are ∼ T 3/2 (Incropera et al., 2007). A 3 K temperature increase will result in an increase in the diffusion coefficient by only a factor of 1.015. The eddy diffusivity characterizes the degree of turbulent mixing in the chamber, which could be increased by temperature variations. When UV lights are turned on, the temperature in the chamber increases by 2-3 K and 15 then gradually decreases to the original temperature from a well-controlled recirculating chilled water system. This small fluctuation is not expected to cause a dramatic increase in turbulence in the chamber. Therefore, increase of vapor wall loss for high UV experiments is deemed incapable of explaining the modeling discrepancy.

Atmospheric implications 20
From the systematic analysis of the experiments and mechanism predictions (and in the absence of explicit autoxidation mechanisms), the most likely hypothesis for the inability of the base case in GECKO-A to simulate both the high and low UV experiments is that later-generation species overcontribute to the SOA in GECKO-A. To compensate for this overcontribution necessitates the assumption of rapid vapor wall loss in order 25 33184 dictions from GECKO-A may then overestimate SOA. If instead all reaction pathways in GECKO-A are realistic but the rates are simply overpredicted, these compounds may still form in the atmosphere but will be affected by competing processes such as dry and wet deposition and uptake into cloud particles (Hodzic et al., 2014;Mouchel-Vallon et al., 2013). If volatilities are too low in GECKO-A, atmospheric predictions would then 10 be biased high. Finally, if higher-generation species simply exhibit a faster wall loss, these compounds will be produced quickly in the atmosphere and will generate significant SOA. An important implication of our analysis is that vapor wall loss in the Caltech chamber is likely slower than that of the base case assumed in GECKO-A. This slower wall loss is 15 consistent with measurements in the Caltech chamber (Loza et al., 2010Zhang et al., 2014) but is considerably slower than those measured in the 8 m 3 chamber used by Yeh and Ziemann (2015), 10 −3 -10 −4 s −1 . Differences in the size and operation of two chambers may contribute to the differences in measured wall loss rates.
The vapor wall loss implied by the present experiments results in an order of mag-20 nitude slower rate than that fit by Zhang et al. (2014) during toluene SOA photooxidation. This difference could potentially indicate differences between chemical systems or could also arise from the manner in which the modeling was performed:  used the Statistical Oxidation Model of Cappa and Wilson (2012) vs. the GECKO-A model used here. An implication of this difference is that the wall loss bias 25 (the ratio of the yield predicted in the absence of vapor wall loss divided by the observed yield) for the α-pinene system is likely lower than the 1.6 predicted by Zhang et al. (2014). However, because we could not definitively fit here both the high and low UV experiments, we cannot calculate a corresponding wall loss bias. Autoxidation is likely important in this system and may potentially explain some or all of the overlap of the growth curves at high and low UV. Explicit mechanisms and rate constants are needed in order to test this hypothesis. If significant, the presence of autoxidation could change the vapor wall loss rate implied by these simulations.
Although other processes do not reconcile the high/low UV model discrepancy, such 5 processes may still be important. Condensed-phase photolysis, though not significant over the short timescale of chamber experiments, has been shown to potentially have a more significant effect in the atmosphere . Particle-phase reactions have been shown to generate significant amounts of high-molecular weight, low volatility species in the α-pinene system (Docherty et al., 2005;Lopez-Hilfiker et al., 10 2015; Gao et al., 2004b), although more efficiently in ozonolysis (Lopez-Hilfiker et al., 2015). Even if particle-phase dimerization is significant in this system, the assumed irreversible formation of dimers in GECKO-A may not accurately represent particlephase reactions, which have been suggested to be reversible (Ziemann and Atkinson, 2012;Trump and Donahue, 2014).

Conclusions References
Tables High UV/OH Figure 3. Schematic illustrating various explanations of the behavior of the SOA growth curves at high and low UV conditions (high vs. low OH). Horizontal lines at the top of each panel denote the chamber wall. Dashed vs. solid and thick vs. thin lines illustrate differences in relative rates for various processes. Panel 1: low rate of vapor wall loss (dashed lines). Dashed lines in low UV reaction sequence indicate slower chemistry due to lower OH levels. Panel 2: later-generation OH reaction rate constants reduced to approximate a lesser contribution of later-generation species to SOA (thinner reaction arrows). Panel 3: condensed-phase photolysis is included, which will be more efficient at high UV. Panel 4: fast vapor wall loss (solid lines). Autoxidation is favored under low UV owing to longer peroxy lifetimes.