To produce a laboratory surrogate of secondary BrC, we generated a mixture of light-absorbing products by the aqueous photo-oxidation of resorcinol and H2O2 exposed to UV-B radiation. Initially, the solution of resorcinol and H2O2 exhibited no absorption of light at visible wavelengths, as shown in Fig. S2. Following 4 h of photo-oxidation, the solution was highly coloured, exhibiting broad absorption features that were strongly dependent on wavelength, such that absorbance across a 1 cm path length was about 0.6 at 400 nm but was negligible at >600 nm (see Fig. S2). In the past, light-absorbing products of the aqueous photo-oxidation of resorcinol were also observed by Chang and Thompson (2010), who used 1 mM H2O2. Here, the concentration of H2O2 was 1 order of magnitude greater, and the colour developed slightly faster, as illustrated by the time series of absorbance at 450 nm (see Fig. S3).

More generally, light-absorbing products of aqueous photo-oxidation have been observed for a wide range of phenolic species, including those with methyl, methoxy, and carbonyl substituents (Chang and Thompson, 2010; Gelencsér et al., 2003; Smith et al., 2016). In detailed mechanistic studies, products of both hydroxylation and oligomerization have been identified (Hoffer et al., 2004; Li et al., 2014; Sun et al., 2010; Yu et al., 2014, 2016). Oligomers form by C–C or C–O radical coupling (Kobayashi and Higashimura, 2003). C–C coupling might be expected to lead to greater absorption enhancement, since the resulting biphenyls and larger oligomers may have some degree of delocalization across the rings (Zhang et al., 2010). Indeed, based on density functional theory calculations, Magalhães et al. (2017) have shown that the absorptivity of bi- and terphenyls is greater in water than in the gas phase, because the planarity and delocalization increase. Interestingly, 5 of the 10 most abundant products of the aqueous photo-oxidation of 0.1 mM syringol with 0.1 mM H2O2 have been shown to be biphenyls (Yu et al., 2014). The concentration of the phenolic precursor used here was 2 orders of magnitude greater, a condition that may favour oligomerization over hydroxylation. Fragmentation may also occur during aqueous photo-oxidation, but the volatile products are removed in the activated carbon denuders following atomization. Based on these considerations, the secondary BrC aerosol is likely composed primarily of oligomers, and we assume the average molecular mass is 326 gmol-1, representative of a terphenyl product of resorcinol. Before investigating the effects of photolysis and heterogeneous OH oxidation on the optical properties of the BrC surrogate, we performed control experiments without either H2O2 or irradiation to constrain the size dependence of the optical properties. Results from a deposition experiment at 60 % RH, in which BrC was injected into the chamber and diluted with clean humidified air, are shown in Fig. 4. The initial geometric mean number diameter and standard deviation were 125 nm and 1.5, respectively, and the initial SSA was about 0.76. When the particles were unperturbed by radiation or OH, they were simply lost by dilution and deposition. Because smaller particles deposit faster than larger ones, the geometric mean surface diameter gradually increased from about 170 to 190 nm over the course of 3 h. Both the absorption and scattering coefficients steadily decreased as particles were continuously lost, but the SSA increased slightly (see Fig. 2b) due to the increase in the geometric mean surface diameter.

This observed size dependence of SSA can be compared to Mie theory calculations, based on the measured size distributions and an assumed complex refractive index. The complex refractive index should not change without photolysis or heterogeneous OH oxidation, so we manually scanned values of n and k to apply for the duration of the experiment. We found that n and k values of 1.35 and 0.046, respectively, are suitable for reproducing the evolution of SSA in this particular experiment (see Fig. 4b). Stepping n from 1.40 to 1.55 and adjusting k accordingly demonstrates that the good agreement for this refractive index is unique, as shown in Fig. S4. We stress that even an approximate value of the refractive index is adequate to distinguish whether changes in the SSA are due to changes in the size distribution or chemical composition. For SOA generated from anthropogenic and biogenic volatile organic compounds, n values ranging from 1.35 to 1.61 have been observed (Kim and Paulson, 2013). The value of k is in the regime of previous observations for BrC at 405 nm, which vary from very low values of 0.004 (Cappa et al., 2012) and 0.007 (Lack et al., 2012) to 0.112 (at 400 nm) (Kirchstetter et al., 2004). In a recent field study in India, a similar value of 0.037 for ambient BrC was observed (Shamjad et al., 2016).

In one experiment, BrC particles were sampled from the chamber, alternately, with and without a diffusion dryer upstream of the DMA. The conditioning was alternated every 15 min, allowing time for the RH of the sheath flow to stabilize. As shown in Fig. S5, the size distribution slowly shifts to larger mobility diameters, because smaller particles are lost faster than larger particles. Besides this trend, the variation between the conditions is negligible, so there is not significant water uptake at 60 % RH.

Having constrained the size dependence of the optical properties, based on the closure between measured and calculated SSA values, we now consider the effects of photolysis on the optical properties of the BrC surrogate. The emission of the UV-B black lights in the chamber, with a peak at 310 nm, is not representative of natural sunlight, so we do not draw direct comparisons between the timescale of our experiments and that of photolysis in the atmosphere. Rather, we performed photolysis experiments to account for the effects of photolysis during the heterogeneous OH oxidation experiments. Wong et al. (2015) have shown that for this chamber equipped with UV-B bulbs, the photon flux at wavelengths below 310 nm is much greater than for a clear-sky summer day; for example, at 300 nm, the photon flux from the bulbs is close to its peak value of about 1×1014 photonss-1cm-2nm-1, while the photon flux outside is negligible. Light at wavelengths below 310 nm is likely driving most of the photochemistry during the photolysis experiments, so we believe that the relative impact of photolysis would be small in the atmosphere, even considering that the ambient OH concentration is lower than in the chamber.

Results from a photolysis experiment at 60 % RH are illustrated in Fig. 5. In each experiment, the value of n was taken as 1.35, as described above; the assumption that n is constant for the duration of the experiment is supported by the previous observation that, for BrC generated from ammonium sulfate and methylglyoxal, n did not change as the reaction proceeded (Tang et al., 2016). Despite closely reproducing the preparation of the BrC surrogate solution and injecting it to approximately the same total particle volume in each experiment, we observed small differences in initial size distributions and SSA values, so it was necessary to vary k slightly from experiment to experiment. For the calculated SSA to follow the trend observed before the lights are turned on, at relative time zero, a k value of 0.040 must be adopted. When strict comparisons are made between experiments, the relative SSA (the current SSA divided by the initial SSA) is presented. In 3 h, the geometric mean surface diameter increased from about 160 to 180 nm, similar to the deposition experiment, so little or no volatilization of the particles can be inferred. Because the diameter increased uniformly, the predicted SSA increases steadily. In contrast, the observed SSA gradually decreased once the lights were turned on, indicating that the particles slowly became more absorbing.

In the past, photolysis has been observed to lead to the absorption enhancement of other BrC surrogates (Saleh et al., 2013; Wong et al., 2017; Zhao et al., 2015; Zhong and Jang, 2014). For example, the absorbance at 400 nm of water-soluble species produced by the combustion of kaoliang stalks has been shown to increase during the first 30 min of irradiation in a solar simulator (Zhao et al., 2015). Recently, the peak absorption enhancement due to the photolysis of biomass burning BrC constituents has been shown to increase with their molecular mass (Wong et al., 2017). Since the trace gas denuders remove volatile species after atomization, the composition of the particles is likely skewed towards an average molecular mass that is larger than that of the species in solution, possibly making the particles more susceptible to photolysis. In general, an absorption enhancement may occur due to direct photolysis of chromophores, followed by radical recombination leading to larger, more absorptive oligomers or reactions with OH from the dissociation of H2O2 generated from the photolysis of carbonyl compounds in the condensed phase (Anastasio et al., 1997; von Sonntag and Schuchmann, 1991). In the past, the absorption enhancement of biomass burning BrC exposed to natural sunlight has been shown to increase with RH (Zhong and Jang, 2014), a trend that was attributed to the increasing production and dissociation of H2O2 with increasing RH. Here, we observe similar trends in relative SSA under both low and high RH conditions (see Fig. 6b). It is unlikely that sufficient OH could be produced in the particle phase at the lower RH, so the indirect photolysis of carbonyl compounds likely does not have a significant role in the observed absorption enhancement.

In past studies, the period of absorption enhancement due to photolysis has been shown to be followed by bleaching (Wong et al., 2017; Zhao et al., 2015; Zhong and Jang, 2014). In the present experiments, the irradiation time is limited by the particle losses; after about 3 h, the scattering and absorption coefficients approach the detection limits of the PASS. Near the end of the photolysis experiments, SSA levelled off, and this period may coincide with the inflection point between absorption enhancement and bleaching.

The evolution of the optical properties of the BrC surrogate during a photo-oxidation experiment is affected by deposition, photolysis, and heterogeneous OH oxidation. Results of a representative photo-oxidation experiment at 60 % RH are shown in Fig. 7; replicate experiments were performed at each RH. The initial geometric mean surface diameter was slightly higher than in the photolysis experiment described above (about 196 nm compared to 160 nm), so although the initial aerosol was slightly more scattering (higher SSA), the value of k required to reproduce the SSA before irradiation is very similar (0.041 compared to 0.040). During irradiation, the geometric mean surface diameter first increased slightly due to deposition and then began to decrease due to volatilization. As a result, the predicted SSA first increases slightly and then, begins to decrease at about 90 min. The observed SSA exhibited a sharp decrease within only 10 min of reaction time and an induction period before a significant increase. Both the absorption enhancement and bleaching contrast with the predicted trend of SSA, so they are certainly due to the evolution of particle composition, rather than size. The absorption enhancement was rapid enough to lead to a peak in the absorption coefficient, despite continuous particle losses (see Fig. S6).

As described above, the evaporation of water from particles can result in a negative bias in photoacoustic measurements (Langridge et al., 2013). If heterogeneous OH oxidation significantly increased the hygroscopicity of the particles, the water content of the aerosol would increase during the experiment. The resulting increase in the magnitude of the bias could contribute to apparent bleaching. In addition to this instrumental artifact, progressively more water uptake could lead to genuinely greater scattering coefficients, which would also result in apparent bleaching. We investigated these potential effects by alternately sampling with and without a diffusion dryer downstream of the chamber at 60 % RH. As shown in Fig. S7, the absorption coefficient does not depend on the conditioning. The scattering coefficient is about 10 % lower for the dried particles (see Fig. S8), but this difference is roughly steady during photo-oxidation. In other words, the particles do not become significantly more hygroscopic, and the changes in absorption and scattering coefficients are indeed due to the chemical evolution of the particles.

To ensure that the sequential absorption enhancement and bleaching described above are distinctive features of the BrC surrogate, we also investigated the heterogeneous OH oxidation of a yellow dye aerosol at 60 % RH. In this case, we observed uniform bleaching, as shown in Fig. S9, consistent with bulk aqueous studies of similar azo dyes (Georgiou et al., 2002). There was no decrease in the geometric mean surface diameter, so there was little or no volatilization. It is likely that OH attacks the azo nitrogen–nitrogen bond (Hisaindee et al., 2013). The product fragments would be large enough to remain in the particle phase; for example, the smaller fragment would still contain one sulfonate group and two aromatic rings.

Unlike photolysis, heterogeneous OH oxidation is strongly dependent on RH, as shown in Fig. 6a. In contrast to the rapid absorption enhancement and bleaching observed at 60 % RH, the period of absorption enhancement is prolonged at 15 % RH, on a timescale similar to that of the photolysis experiments. The peak absorption enhancement at 15 % RH results in a lower relative SSA than at 60 % RH. To better understand the effects of RH on heterogeneous OH oxidation, we compare experimental results with those of the multi-layer kinetics model described above. In all, the model has five adjustable parameters; these are two uptake coefficients, γOH,i, the diffusion coefficient, Db,i, and the molar absorptivities of products B and C, εB, and εC, respectively.

At 60 % RH, the experimental features can be reproduced (see Fig. 8) by setting γOH,A and γOH,B to 5.0 and 7.5, respectively, suggesting that oxidation is very efficient. An uptake coefficient, γOH,A, greater than unity indicates that each collision between OH and A leads to the formation of more than one molecule of B. In previous heterogeneous OH oxidation experiments, a wide range of OH uptake coefficients have been observed, including values greater than unity (George and Abbatt, 2010). In the particle phase, oxidation initiated by OH may form products directly from reactions with OH or indirectly from organic radicals, RO. The organic radicals may result from free-radical chain reactions that have been shown to lead to very high effective uptake coefficients, in particular, when species like NO and SO2 are present to enhance the conversion of RO2 to RO (Richards-Henderson et al., 2015, 2016). We note that the values of γOH,i are dependent on the assumed average molecular mass, which is set to 326 gmol-1, representative of a terphenyl oligomerization product of resorcinol, as discussed above. A greater molecular mass would result in lower values of γOH,i, so we cannot be sure to what extent free-radical chain reactions are playing a role in the experiments. Scaling the uptake coefficients such that γOH,A decreases from 5.0 to 1.0 requires increasing the molecular weight from 326 to 1600 gmol-1. It is difficult to rationalize such large oligomers forming during the 4 h aqueous photo-oxidation, so secondary radicals are indeed likely contributing to oxidation. Furthermore, we assume that OH undergoes solely gas-surface reactions, following Shiraiwa et al. (2009). Recent measurements indicate that this may be an over-simplification, as the reactive–diffusive length of OH in squalene is similar to that of ozone (Lee and Wilson, 2016), so surface-layer reactions may also occur. Nonetheless, we can state with confidence that the oxidation is very efficient.

We find that the observed trend in relative absorption at 60 % RH – in particular, the significant bleaching – can be reproduced only if the aging particles are taken as well-mixed, so the diffusion of A from the bulk to the surface is not restricted. As shown in Fig. 8b, the concentrations of A, B, and C are the same in all four layers. At 60 % RH, secondary organic material derived from the gas-phase photo-oxidation of toluene has a diffusion coefficient on the order of 10-10 cm2s-1 (Song et al., 2016). Here, any value greater than 1×10-14 cm2s-1 gives good agreement, so Db,i is set to 1×10-14 cm2s-1 as a lower limit. Though the extent of bleaching is captured by the model, rapid initial colour enhancement is not. We speculate that the other processes, like auto-oxidation, could contribute to this feature. The same diffusion coefficient is assumed for all three species, but in reality Db,i may increase in going from A to B and decrease in going from B to C; if C is a product of fragmentation, it might be expected to have a greater diffusion coefficient in A than the self-diffusion coefficient of A. Indeed, plasticization has been shown to occur during heterogeneous OH oxidation of semi-solid alkane particles at high gas-phase OH concentrations (Wiegel et al., 2017).

Finally, the experimental features are reproduced by setting the molar absorptivities of B and C (εB and εC) to 1.5εA and zero, respectively. We note that this combination of parameters is likely one of several possible solutions. Even for a more tightly constrained set of experimental data regarding consumption of oleic acid by heterogeneous reactions with ozone (Ziemann, 2005), more than one model scenario has been found to adequately reproduce observations of the number of oleic acid molecules in the particle phase (Pfrang et al., 2010).

At 15 % RH, the experimental features are reproduced (see Fig. 8) by setting γOH,A and γOH,B to 1.0 and 0.1, respectively, and Db,i to 1×10-16 cm2s-1. The decreases in the uptake coefficients with RH may be the result of differences in the distribution of initial BrC species in the particle or the orientation of the species at the surface. The decrease in Db,i by 2 orders of magnitude is reasonable, considering the significant RH dependence on viscosity observed for secondary organic material derived from toluene and isoprene (Song et al., 2015, 2016). The diffusion coefficient of secondary organic material derived from toluene was found to be less than 10-17 cm2s-1 for RH≤17% (Song et al., 2016). As before, we set εB and εC to 1.5εA and zero. It is possible that fragmentation products volatilize out of the condensed phase to a greater extent than they diffuse to the bulk phase, because the particles are so viscous. In fact, a slight decrease in the mean geometric surface diameter suggests that there is some degree of volatilization. On a related note, because the viscosity is so high, the products B and C are much more concentrated at the surface than in the bulk layers (see Fig. 8c). Consequently, at 15 % RH, the aging particles likely consist of less absorptive cores and highly absorptive shells.

In general, we emphasize that the set of molecular properties presented above, which best fit the experimental data, were selected after multiple trials with different input parameter values and different model scenarios as well. Although the mechanism and final parameter set fit the data remarkably well, we acknowledge that the complex nature of the inherent chemistry, with changes in optical properties occurring alongside concentration, suggests that this parameter set should only be viewed semi-quantitatively; i.e. the solution is most useful for substantiating the mechanism leading to the changes in absorption.