Toluene (EMD Chemicals, ≥ 99.8 %) and m-xylene (Sigma Aldrich, ≥ 99 %) were continuously injected by a syringe pump (Chemyx, Fusion 200) into a glass round-bottom flask (Ace glass, 100 mL). The flask was held at 310 K, and the organic liquids vaporized at the tip of a syringe prior to the formation of a falling drop. The resulting gas-phase molecules were swept in a flow of pure air (Aadco 737 Pure Air Generator; 1.0 L min-1) into an oxidation flow reactor (OFR). The injected precursor concentration was 5.0 ± 0.5 ppm, which was calculated from the injection rate of the liquid into the flow rate of the reactor.

Aerosol particles composed of secondary organic material were produced in the oxidation flow reactor (Kang et al., 2007; Lambe et al., 2011). Aromatic precursors were oxidized by hydroxyl radicals (OH) (Fig. 1), and some of the resulting low-volatility products contributed to new particle production in the OFR. The experimental conditions are listed in Table 1. The OFR was operated at a temperature of 293 ± 2 K, a flow rate of 7.0 ± 0.1 L min-1, and a residence time of 110 ± 2 s. Hydroxyl radicals were produced by photochemical reactions involving ozone and water inside the OFR as follows: (i) O3+hν (254 nm) → O2+ O(1D) followed by (ii) O(1D) + H2O →2 OH. Ozone was generated outside the reactor by irradiating pure air with the ultraviolet emissions of a mercury lamp (λ=185 nm). The injected ozone concentration was 13 ± 2 ppm. Water vapor was introduced by bubbling ultrapure water (18.2 MΩ-cm) with air. The relative humidity inside the reactor was 13 ± 3 %.

The experiments were conducted for several different concentrations of NOx. For “low NOx” experiments, no NOx was added, and the nitric oxide (NO) concentration in the pure air was below the detection limit of the NOx analyzer (Eco Physics CLD 899 Y; < 70 ppt NOx). For “high NOx” experiments, NO of 2.5 to 10 ppm was injected using a mass flow controller (Table 1). In a control experiment to assess the possible importance of ozonolysis, in the dark the produced particle mass concentration was 0.1 % of that obtained when the ultraviolet lamps were illuminated, indicating that photooxidation was the major pathway of SOM production.

Spectroscopic ellipsometry of SOM films followed the procedures introduced in Liu et al. (2013). Thin, continuous, mirror-like films of SOM were synthesized by electrostatic deposition of aerosol particles onto silicon substrates. Information about film preparation and characterization was provided in Liu et al. (2013). Spectroscopic ellipsometry was carried out across 280 to 1200 nm at different incident angles using a variable-angle spectroscopic ellipsometer (J.A. Woollam VASE). From the data sets, film thickness, film non-uniformity, and wavelength-dependent real refractive indices, represented by n, were retrieved using the WVASE32 software package (J.A. Woollam VASE). Individual film thickness ranged from 100 to 350 nm. The non-uniformity in film thickness was 5 to 10 % over the ellipsometric sampling spot. An absence of systematic error in retrieved refractive indices, as related to film thickness or film non-uniformity, was confirmed previously (Liu et al., 2013).

For each thin film sample, duplicate measurements were conducted at two or more different spots. The retrieved n values agreed with each other within an absolute difference of 0.015. An overall uncertainty of n, taking both the reproducibility and the fitting error into account, was within ±0.015 (10 and 90 % confidence interval) across the studied wavelength range from 280 to 1200 nm. As a confirmation of the overall approach, the n values retrieved for squalane, a non-absorbing standard, and nigrosin dye, a light-absorbing standard, were consistent with literature values (cf. supporting information in Liu et al., 2013).

Particles were collected onto Teflon filters (Millipore FGLP, 0.2 µm pore size). The collected SOM mass was determined by weighing the filter before and after sample collection using an analytical balance (Sartorius, LA120S; 2.0 to 10.0 mg, with an uncertainty of 0.1 mg). The filters were extracted in methanol (40.0 to 100.0 mL) while ultrasonicating (Branson 2510; 20 min). The concentration c was calculated from the measured SOM mass and the volume of the solvent. An extraction efficiency of 100 % was assumed (Updyke et al., 2012). The extract was pipetted into a quartz cuvette having an optical length of 10 mm. Absorbance spectra were recording using an ultraviolet–visible spectrometer (Agilent Model 8453). The spectrum of neat methanol was used as baseline. Analysis of blank filters showed no absorbance across the studied region of 240 to 800 nm.

The imaginary refractive indices k were calculated from the data sets as follows (Sun et al., 2007): k=ln⁡(10)4πρλcLA(λ) for an absorbance A, an optical path length L (m), a material density ρ (kg m-3), and a concentration c (kg m-3). A material density of (1.4 ± 0.1) × 103 kg m-3 was used in the analysis for toluene- and m-xylene-derived SOMs (Ng et al., 2007). For Suwannee River fulvic acid (International Humic Substances Society, 2S101F), a material density of 1.3 × 103 kg m-3 was used. For the k values derived from UV–visible spectroscopy, propagation of uncertainties in ρ, c, and A(λ) leads to an overall uncertainty of ±15 % (10 and 90 % confidence interval) for λ < 420 nm.

In an alternative to the UV–visible spectroscopy, the k values are also retrieved by ellipsometry (Liu et al., 2013). A comparison of k values derived from UV–visible spectroscopy and to those from ellipsometry is provided in the supporting information (cf. Fig. S1 in the Supplement). Overall agreement is good. For the smallest k values (< 0.005), the ellipsometry retrievals have uncertainties approaching > 50 %. The uncertainties for UV–visible spectroscopy are considerably smaller (< 15 %). Even so, this method requires sample extraction, and artifacts associated with extraction efficiency, material density, and solvent effect can be introduced. All factors considered, the k values derived from UV–visible spectroscopy were adopted in this study for further analysis.

Aerosol particles were collected on Teflon filters (Sartorius Stem, 0.2 µm) at a flow rate of 2 L min-1 for up to 24 h. The collected mass on the filters ranged from 0.8 to 2.0 mg. The Teflon filter was cut to the shape of the germanium element of an attenuated total reflectance (ATR) accessory (Pike Technologies). The assembly was then screw-pressed to the crystal surface, the holder was opened, and the filter was peeled off. A thin layer of secondary organic material remained on the surface of the crystal (Hung et al., 2012). An experiment using a blank filter showed no residual signal from the Teflon filter after peeling. This preparation method avoided interference from Teflon filters across 900–1250 cm-1 (Russell et al., 2009b, 2011; Takahama et al., 2012), which otherwise obscured the absorption bands of C–O stretching.

After preparation, the filter samples were taken for spectroscopy analysis. Infrared spectra were recorded using the ATR accessory in a Fourier transform infrared spectrometer (FTIR, Nicolet 670). The spectral resolution was 0.5 cm-1. The number of scans was 16. Additional information about the ATR-FTIR protocols is provided in Hung et al. (2012). A band fitting algorithm, implemented in MATLAB, was used to analyze the infrared spectra. The algorithm was adopted from Russell et al. (2009a) and Takahama et al. (2012). The absorption bands of alkanes (C–H), carboxylic hydroxyls (O–H), alcoholic hydroxyls (O–H), and carbonyls (C = O) were identified using the literature-described algorithm. The algorithm was further developed in the present study to characterize nitrate (-ONO2), nitro (-NO2), and ether (C–O–C) groups.

The wavelength-dependent absorptive component k of the refractive index is plotted in Fig. 2a for toluene- and m-xylene-derived SOMs prepared at several different initial NO concentrations (cf. Table S1 in the Supplement for tabulated data). The k values increase with increasing initial NO concentration. For toluene-derived SOMs, the k values at 405 nm range from 0.0017 to 0.0153. These values compare to a range of 0.0018 to 0.0072 reported by Nakayama et al. (2013) for SOMs produced by toluene photooxidation in an environmental chamber.

The increase of k for high NOx can in part be explained by the production of light-absorptive organonitrogen compounds, mostly nitro-aromatic compounds, such as nitrophenols, nitrocatechols, and dinitrophenols (cf. Sect. 3.2). These compounds have been identified as products from toluene photooxidation under high-NOx conditions (Forstner et al., 1997; Jang and Kamens, 2001b; Sato et al., 2007; Zhong et al., 2012; Nakayama et al., 2013). Nitro-aromatic compounds have also been identified in brown carbon sampled in urban plumes dominated by anthropogenic SOM (Zhang et al., 2011, 2013). The spectra for several methyl-nitrophenol isomers were measured (cf. Sect. S1 in the Supplement), and the results confirm that these compounds are strong UV absorbers. In particular, the aromatic compounds having hydroxyl and nitro groups in para substitution, such as 2-methyl-4-nitrophenol (a major product of aromatic photooxidation), have a strong absorption band at 320 nm. Compounds having this configuration are good candidates for contribution to the main peak in the difference spectra Δk (i.e., Δk=k-kNO0=0) (Fig. 2b).

For similar reaction conditions, the m-xylene-derived SOMs are less absorptive than the toluene-derived SOMs (Fig. 3a; Table 1). The filter samples of toluene-derived SOMs have a yellowish to light brownish color, even for samples collected in low-NOx experiments. For comparison, m-xylene samples have a light yellowish color only for high-NOx experiments and are visually white otherwise.

The k values of the aromatic-derived SOMs can be compared to those of other light-absorbing material relevant to atmospheric aerosol particles (Fig. 4). The k values decrease for increasing wavelength for the aromatic-derived SOMs. Similar wavelength-dependent behavior is observed for light-absorbing carbonaceous materials referred to as “brown carbon” in literature (Kirchstetter et al., 2004; Andreae and Gelencser, 2006; Hoffer et al., 2006; Alexander et al., 2008; Dinar et al., 2008; Chakrabarty et al., 2010; Cappa et al., 2012; Lack et al., 2013). In contradistinction, the value of k for black carbon is independent of wavelength (Kirchstetter et al., 2004). Compared to the k values of SOMs derived from examples of biogenic precursors (B-SOM), such as α-pinene and limonene SOM (Liu et al., 2013), the k values of the studied anthropogenic SOMs (A-SOM) are 1 order of magnitude more absorptive in the UV–visible region, even for those produced at low NOx. These higher values suggest that conjugated double bonds are retained in some oxidation products, which have absorption transitions in the ultraviolet to near visible (Lambe et al., 2013). Even so, the k values in the low-NOx experiments are smaller than those of a reference compound like Suwannee River fulvic acid, which is often cited as a surrogate of atmospheric humic-like substances (HULIS) (Gelencsér et al., 2003; Dinar et al., 2006). In the high-NOx experiments, however, the k values are within the range of atmospheric brown carbon (cf. shaded region in Fig. 4).

The real refractive indices n of toluene- and m-xylene-derived SOMs are shown in Fig. 5 for several different initial NO concentrations. The n values decrease for increasing wavelength. The curves can be parameterized by the three-term form of Cauchy's equation (cf. Table S2). The Cauchy-form of the curves for the studied anthropogenic SOMs also holds for the biogenic SOMs reported previously (Liu et al., 2013).

The refractive indices n shift +0.02 for both toluene- and m-xylene-derived SOMs for an increase of the initial NO concentration from 0 to 10 ppm (Fig. 5). This upward shift of n for increasing initial NO concentration is possibly attributed to an increasing abundance of nitrogen in the SOM produced at higher initial NO concentrations. Nitrogen has a higher atomic polarizability (1.03 Å3) than both oxygen (0.57 Å3) and hydrogen (0.17 Å3) (Bosque and Sales, 2002). The n value of a material is related to its polarizability by the Lorentz–Lorenz equation (Bosque and Sales, 2002). Within the tolerance of the measurement uncertainty, the n values do not differ between SOMs derived from the two different aromatic precursors at the same initial NOx concentration. The implication could be that the n values of SOMs are mainly determined by bulk chemical properties, such as the elemental ratios or functional groups (cf. Sects. 3.2 and 3.3). Detailed chemical properties, such as the molecular structure, might play a minor role in determining the value of n. In this case, upscaling of the laboratory parameterizations to large-scale models of the effects of different types of SOMs on radiative forcing and climate is simplified (Lambe et al., 2013; Flores et al., 2014; Kim et al., 2014). As a caveat, the SOMs of this study were produced at mass concentrations much higher than typical atmospheric concentrations. Both elemental composition and refractive index can depend on mass concentration (Shilling et al., 2009; Kim et al., 2012; Kim and Paulson, 2013), and further investigations are needed to quantify this possible effect.

The infrared spectra in the presence and absence of NOx are similar, except for organonitrogen groups, such as -NO2 and -ONO2 (Fig. 6). This similarity suggests that the oxygen-containing functional groups, excluding nitrogen-containing groups, are substantially similar for SOMs produced at the different NOx concentrations. SOMs derived from toluene and m-xylene also have similar overall compositions.

Organonitrogen compounds are detected in SOM collected in high-NOx experiments. The bands at 846, 1281, and 1647 cm-1, corresponding to organonitrate groups (-ONO2), are present in both toluene- and m-xylene-derived SOMs (Roberts, 1990; Liu et al., 2012). The area of the -ONO2 band from 1610 to 1690 cm-1, when normalized by the alkane C–H bands from 2790 to 2980 cm-1 to account for different masses on the filters, increases for greater initial NO concentrations (Fig. 3c). Comparison of the spectrum of the toluene-derived SOM to that of m-xylene-derived SOM for fixed initial NOx concentration shows that the -ONO2 fractions are approximately equal for both types of SOMs produced. The dominant mechanism for -ONO2 production, which is the reaction of peroxy radicals (RO2) with NO (Roberts, 1990), can explain why the fraction of -ONO2 increases for greater NO concentrations.

In the high-NOx experiments, -NO2 groups are produced. For the toluene-derived SOM, the production of -NO2 groups is indicated by a strong band at 1558 cm-1 and a weak band at 1342 cm-1 (Fig. 6a). For m-xylene-derived SOM, only the strong band at 1558 cm-1 is observed (Fig. 6b). Based on an analysis of area ratios, the mole fraction of -NO2 groups in the m-xylene-derived SOM is 35 to 50 % lower than that in the toluene-derived SOM for fixed initial NOx concentration (Fig. 3c). The production mechanism of -NO2 group has been proposed as the adduction of -NO2 to phenoxy radicals to produce nitrophenols (Forstner et al., 1997; Jang and Kamens, 2001b; Nakayama et al., 2013). Compared to toluene, the alkyl substitution at a metasite of m-xylene can inhibit the production of stable -NO2 adducts of phenoxy radicals (Nakayama et al., 2013), which can explain the lower -NO2 fraction observed for m-xylene-derived SOM.

The difference in -NO2 fraction explains in part but not entirely the differences in k values for toluene- compared to m-xylene-derived SOMs (cf. Sect. 3.1). When normalized by -NO2 fractions and for similar reaction conditions, Δk at 320 nm for toluene-derived SOM is 50 % higher than that of m-xylene-derived SOM. The implication is that the compounds in toluene-derived SOM are more UV-absorptive than those in m-xylene-derived SOM. Differences in the extent of conjugation of the oxygenated products can be important. The k values in the UV region at low NOx (kNO0=0) provide a baseline to quantify this influence. When normalized by both kNO0=0 and -NO2 fraction, the two types of SOM are similarly absorptive (Fig. 3d). Organonitrogen groups attached to a conjugated chain can have increased light absorption as well as shifts in absorption to longer wavelengths.

The infrared spectra show that several different types of oxygenated functional groups are present in the SOMs (Fig. 6). The groups include alcoholic hydroxyls (O–H) at 3100–3700 cm-1, carboxylic hydroxyls (O–H) at 2400–3300 cm-1, carboxylic carbonyls (C = O), and mixed ketones / aldehydes (C = O) at 1640–1850 cm-1. The spectra of aromatic-derived SOMs are similar to those reported in the literature for related SOMs (Jang and Kamens, 2001a; Liu et al., 2012). The new finding of the present study is the presence of a strong C–O stretch at 1000–1260 cm-1 (cf. Table S3). This band was obscured in previous studies by ammonium sulfate or Teflon filter. As explained for Fig. 7, this C–O band cannot be explained by alcohols, phenols, cyclic anhydrides, carboxylic acids, or other carbonyls that have been identified as major products from oxidation reactions of aromatic precursors.

Figure 7 shows a ternary diagram representing the relative areas of O–H, C–O, and C = O bands for toluene- and m-xylene-derived SOMs (cf. Sect. S2). Reference compounds having different types of oxygenated functional group are also plotted. The ternary diagram groups different types of oxygenated compounds into clusters. The individual products from photooxidation of toluene identified by gas chromatography/mass spectrometry (GC/MS) are plotted for comparison (Forstner et al., 1997). The cluster representing the toluene- and m-xylene-derived SOMs is uniquely situated and differentiated from the reference compounds because of the C–O stretch at 1000–1260 cm-1.

This absorption band at 1000–1260 cm-1 is plausibly contributed by an ether group (C–O–C) of acetals and hemiacetals produced via particle-phase reactions (Jang et al., 2002; Kroll and Seinfeld, 2008; Lim et al., 2010). These reactions tend to drive product distribution toward the C–O vertex of the composition diagram (cf. arrow in Fig. 7). The gas-phase oxidation of aromatic precursors produces dialdehydes in high yields, including glyoxal and methylglyoxal. These dialdehydes readily oligomerize along hemiacetal and acetal pathways, with associated changes in the C–O / C = O stretch band ratio (Loeffler et al., 2006). Hemiacetal / acetal production reactions leading to oligomerization can occur in SOM produced by photooxidation of trimethylbenzene (TMB), even in the absence of catalysis by sulfuric acid (Kalberer et al., 2004).

The mole fraction of each functional group is estimated using the absorptivity of Russell et al. (2009b) and Takahama et al. (2012), along with the area ratios of ether (C–O–C) to alkane (C–H) bands for 19 ether and acetal compounds appearing in the NIST (National Institute of Standards and Technology) database. The analysis concludes that ether groups constitute up to 50 % of the SOM mass. This result agrees with a modeling study suggesting that 20–80 % of the SOM derived from toluene is produced by particle-phase reactions (Cao and Jang, 2009). These particle-phase reactions can produce oligomers having conjugated structures that contribute to the light absorption even in the absence of nitrogen moieties (Zhong et al., 2012).

For the obtained spectral data sets of n and k (Sect. 3.1), the optical effects of brown carbon (BrC) from anthropogenic SOM can be assessed. A model case study is formulated to represent light-absorbing particles in a fresh urban plume close to the anthropogenic sources. Parameters defining the case study are listed in Table 2. The case study considers a population of brown carbon particles produced by photooxidation of anthropogenic aromatic precursors in the presence of NOx. This population is compared with populations of black carbon (BC) particles (representing emissions from fossil fuel combustion) and externally mixed at variable ratios with ammonium sulfate particles (representing the regional background atmospheric aerosol). The number–diameter distributions of the BrC and sulfate particle populations are representative of polluted urban regions (Wu et al., 2008). The BC particle population having a relatively smaller mode diameter is typical for fresh soot particles emitted from motor vehicles (Kleeman et al., 2000). The investigated ratios of BC and ammonium sulfate are representative of Asian outflows (Ramana et al., 2010). The external mixing assumption is consistent with the small absorption enhancement of BC in urban regions (Cappa et al., 2012). The single-scattering albedo ω, defined as the ratio of scattering to total extinction, is calculated for each population and their mixtures using a Mie-theory-based optical model (Bohren and Huffman, 1983; Liu et al., 2013) (Fig. 8a). The relative contribution of BrC absorption to total light absorption (i.e., BrC / (BrC + BC)) is calculated as a function of the mass ratio of organic matter to BC. The calculated results for λ=320, 405, and 550 nm are plotted in Fig. 8b. These three wavelengths are selected because solar radiation in these bands respectively regulates O3 photolysis, NO2 photolysis, and energy balance (cf. Fig. 8a).

Results of the case study have several implications for climate and atmospheric chemistry modeling. The ω values of the BrC particle populations are close to unity for λ > 500 nm (Fig. 8a). When externally mixed with BC, the studied BrC has a negligible contribution to light absorption at 550 nm (Fig. 8b). These results indicate that these BrC populations have a net cooling effect. The ω values, however, decrease below unity for λ < 400 nm, meaning that the particle population becomes absorptive in the UV region. Although the solar irradiance in the UV region contributes only 10 % of the total solar irradiation, meaning a small heating effect by brown carbon for the conditions of the case study, the effect can still be important because UV irradiance determines the photolysis rates of many chemical species (Fig. 8a). For example, reduced UV irradiance for λ < 320 nm slows ozone photolysis, thus suppressing the production of OH radicals (Martin et al., 2003; Tie et al., 2003). The case study suggests that BrC populations can have a substantial contribution for light absorption in this band (Fig. 8b). For a mass ratio of organic matter to BC in a range of 2 to 20, which is typical for urban atmosphere (Turpin et al., 1991), BrC accounts for 15–80 % of the UV absorption at 320 nm. The photolysis of NO2 is similarly suppressed by reduced UV irradiance for λ < 405 nm, thus inhibiting the production of ozone (Dickerson et al., 1997; Martin et al., 2003). The implication is that, as an effective UV absorber, BrC influences the production of O3 and OH by reducing UV irradiance and consequently affects the oxidation capacity of the regional atmosphere.

In conclusion, photooxidation of toluene and m-xylene in the presence of NOx can produce SOMs having k values similar to those reported for brown carbon in biomass burning and urban plumes (Kirchstetter et al., 2004; Hoffer et al., 2006; Alexander et al., 2008; Dinar et al., 2008; Chakrabarty et al., 2010; Cappa et al., 2012; Lack et al., 2013). The implication is that the photooxidation of anthropogenic precursors can be a significant source of atmospheric brown carbon. These findings are consistent with atmospheric observations in urban regions, such as the Los Angeles basin (Zhang et al., 2011, 2013; Cappa et al., 2012), Mexico City (Barnard et al., 2008), and Beijing (Cheng et al., 2011). The case studies considered in the present study suggest that anthropogenic brown carbon, along with brown carbon from biomass burning, can have a major influence on light absorption at wavelengths that drive photochemical reactions. This effect needs to be evaluated in the future modeling studies of atmospheric chemistry.