Oxidative aging of mixed oleic acid/sodium chloride aerosol particles

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2012jd018163, 2012 Oxidative aging of mixed oleic acid/sodium chloride aerosol particles Benjamin J. Dennis-Smither, 1 Rachael E. H. Miles, 1 and Jonathan P. Reid 1 Received 22 May 2014; revised 14 August 2012; accepted 12 September 2012; published 24 October [1] Studies of the oxidative aging of single mixed component aerosol particles formed from oleic acid (OL) and sodium chloride over a range of relative humidities (RH) and ozone concentrations by aerosol optical tweezers are reported. The rate of loss of OL and changes in the organic phase volume are directly measured, comparing particles with effloresced and deliquesced inorganic seeds. The kinetics of the OL loss are analyzed and the value of the reactive uptake coefficient of ozone by OL is compared to previous studies. The reaction of OL is accompanied by a decrease in the particle volume, consistent with the evaporation of semivolatile products over a time scale of tens of thousands of seconds. Measurements of the change in the organic phase volume allow the branching ratio to involatile components to be estimated; between 50 and 85% of the initial organic volume remains involatile, depending on ozone concentration. The refractive index (RI) of the organic phase increases during and after evaporation of volatile products, consistent with aging followed by a slow restructuring in particle morphology. The hygroscopicity of the particle and kinetics of the response of the organic phase to changes in RH are investigated. Both size and RI of unoxidized and oxidized particles respond promptly to RH changes with values of the RI consistent with linear mixing rules. Such studies of the simultaneous changes in composition and size of mixed component aerosol provide valuable data for benchmarking kinetic models of heterogeneous atmospheric aging. Citation: Dennis-Smither, B. J., R. E. H. Miles, and J. P. Reid (2012), Oxidative aging of mixed oleic acid/sodium chloride aerosol particles, J. Geophys. Res., 117,, doi: /2012jd Introduction [2] Atmospheric aerosol particles are typically internal mixtures of a wide range of organic and inorganic components [Pöschl, 2005]. The physical state and chemical composition of atmospheric aerosol particles affects their optical properties [Lang-Yona et al., 2010] and effectiveness as cloud condensation nuclei [Engelhart et al., 2008; Huff Hartz et al., 2005; Pöschl, 2011; Sun and Ariya, 2006], and hence their influence on the radiative balance of the Earth s atmosphere [McFiggans et al., 2006]. The physical state of an aerosol particle is dependent on composition, temperature, and relative humidity (RH) [Cappa et al., 2008b; Martin, 2000]. Aerosol composition can be altered by oxidative aging by gas phase oxidants, such as the hydroxyl radical (OH), nitrate radical (NO 3 ) and ozone (O 3 ), leading to changes in physical state as well as composition [Hung and Tang, 2010; Lee et al., 2012]. The oxidation of organic components leads to new chemical functionality, the formation of new compounds with either reduced or increased 1 School of Chemistry, University of Bristol, Bristol, UK. Corresponding author: J. P. Reid, School of Chemistry, University of Bristol, Bristol BS8 1TS, UK. ( j.p.reid@bristol.ac.uk) American Geophysical Union. All Rights Reserved /12/2012JD volatility [Kroll and Seinfeld, 2008], changes in hygroscopicity [Cappa et al., 2011; Chang et al., 2010; Dennis-Smither et al., 2012; Massoli et al., 2010; Vesna et al., 2008], and changes in optical properties [Cappa et al., 2011], such as in the formation of light absorbing organic components [Gelencsér et al., 2003]. The heterogeneous reaction between gas phase oxidants and organic aerosol also acts a sink for gas phase oxidants [Ravishankara, 1997]. [3] In this publication we study the oxidative aging of single mixed component aerosol particles formed from an inorganic component, sodium chloride (NaCl), and an unsaturated fatty acid, oleic acid (cis-9-octadecenoic acid; OL). NaCl is a significant component of atmospheric aerosol, particularly in marine environments [O Dowd and de Leeuw, 2007], and a component with well-established phase behavior, hygroscopicity and optical properties [Hargreaves et al., 2010; Tang et al., 1977]. Fatty acids such as OL are ubiquitous in the atmosphere and are a component of marine, rural and urban aerosol [Cheng et al., 2004; Mochida et al., 2002; Wang and Kawamura, 2005]. The unsaturated carboncarbon double bond in OL is susceptible to oxidation by ozone leading to the formation of an ozonide. The heterogeneous oxidative aging of OL by ozone has been the subject of several experimental studies and the OL-O 3 heterogeneous reaction system (HRS) has emerged as a benchmark system for the study of the ozonolysis of unsaturated organic compounds. However, the influences of an inorganic seed and RH have 1of13

2 received much less attention [Vesna et al., 2009]. In this paper, we resolve with high accuracy the branching between involatile and volatile products, the changes in hygroscopicity and kinetics of hygroscopic response that accompany oxidation, the change in refractive index due to compositional change, and the separation of time scales for chemical transformation and volatilization of products. We also present preliminary measurements of the kinetics of oxidative processing when the inorganic seed is deliquesced, comparing the rate of disappearance of OL with the case of an effloresced seed. [4] A number of previous studies have identified and reported yields of the products from the OL-O 3 HRS [Hearn and Smith, 2004; Hung et al., 2005; Katrib et al., 2004; Vesna et al., 2009; Ziemann, 2005]. The ozonide formed rapidly decomposes into C 9 fragments. Due to the asymmetry of the molecule about the double bond, this leads to the branching of the reaction scheme into two separate pathways. One gives nonanal (NL) and a Criegee intermediate (CI) with a carboxylic acid end group. The second pathway gives 9-oxononanoic acid (ON) and another CI with an alkyl tail. In the condensed phase, CIs can be rapidly relaxed from their highly excited vibrational states to form stabilized CIs opening the pathway to secondary chemistry through further reactions by these stabilized CI with other compounds present in the solution [Zahardis and Petrucci, 2007; Ziemann, 2005]. [5] There are four commonly observed products: NL and ON which, as mentioned earlier, can form directly from the decomposition of the primary ozonide; and azelaic acid (AA) and nonanoic acid (NA). AA and NA can form either directly by rearrangement of the vibrationally excited CIs or from decomposition of the products of the secondary chemistry of the stabilized CIs. The formation of AA and NA by rearrangement of the excited CIs constitutes only a minor reaction pathway due to the rapid relaxation of CIs in solution and they are more likely products of the decomposition of the secondary chemistry products [Mochida et al., 2006; Zahardis and Petrucci, 2007]. These four products have been directly observed by a variety of methodologies (see Zahardis and Petrucci [2007] for a review of the methodologies employed). [6] Vesna et al. [2009] extended these works by using an aerosol flow reactor and off-line gas chromatography mass spectrometry to provide quantitative yields of the four major C 9 ozonolysis products (NL, ON, AA and NA) in both the gaseous and condensed phases. From the four, NL and ON were found to have the highest yields, which might be expected as they form directly from decomposition of the primary ozonide. Vesna et al. [2009] also investigated the effect of RH on the product distribution and did not observe a systematic change. Previous studies of the OL-O 3 HRS have also observed the formation of high molecular weight oligomeric products [Katrib et al., 2004; Last et al., 2009; Vesna et al., 2009; Ziemann, 2005]. These products form from the secondary chemistry between the CIs and organic compounds, forming polymeric structures consisting of C 9 units linked by peroxide, ozonide or ester bonds. Ziemann [2005] estimated 68% of the condensed mass to be formed from such compounds. [7] We have previously used aerosol optical tweezers to study the phase, morphology and hygroscopicity of mixed NaCl/OL particles during oxidative aging by ozone [Dennis-Smither et al., 2012]. We used bright field microscopy and cavity enhanced Raman spectroscopy (CERS) to investigate the structure of NaCl/OL particles in which the inorganic phase is either effloresced or deliquesced. In deliquesced NaCl/OL particles, a phase-separated morphology is adopted consisting of two partially engulfing liquid phases that can be identified as hydrophobic and hydrophilic phase domains, consistent with the equilibrium morphology predicted from the surface and interfacial energies [Reid et al., 2011]. The CERS spectra of NaCl/OL particles in which the inorganic phase is effloresced are consistent with a NaCl crystalline inclusion within a spherical OL host droplet. [8] In this publication we explore the oxidative aging by ozone of both effloresced and deliquesced NaCl/OL particles, focusing on the reaction kinetics and the branching observed between volatile and involatile products. To achieve this, we utilize the high level of accuracy in estimating droplet size that can be attained from the CERS fingerprint. Such detailed data should prove of considerable value when benchmarking kinetic models of heterogeneous aging. Indeed, composition changes can be followed from changes in the spontaneous Raman band intensities, and changes in particle size and refractive index can be estimated from the CERS fingerprint during the oxidation reaction and subsequent evaporation of products. The homogeneity of the particle phase can also be assessed. Section 2 describes the experimental strategy, results are presented and discussed in section 3, and section 4 presents our conclusions. 2. Experimental Strategy [9] We have described the use of optical tweezers to isolate and probe the phase, morphology and change in hygroscopicity of mixed NaCl/OL particles during oxidative aging by ozone in a previous publication [Dennis-Smither et al., 2012]. In this section we will review the approach. [10] Particles are held within a trapping cell, the inner walls of which are made of Teflon, using a tightly focused light beam from a 532 nm laser. Particles are imaged using conventional bright field microscopy and probed using CERS. The gaseous environment is controlled by the flow of a gas stream through the cell with a total flow rate up to 0.35 L min 1, regulated by mass flow controllers. The RH is controlled by mixing dry and humidified nitrogen, with the latter produced by passing a nitrogen stream through a water bubbler. The RH and temperature of the gas flow are measured prior to entering the trapping cell (HUMICAP HMT 331, Vaisala). The particles can be exposed to ozone produced by an ozone generator (Model 600, Jelight) consisting of an UV light source through which a flow of nitrogen and synthetic air is passed. The ozone concentration can be varied by controlling the ratio of synthetic air to nitrogen and is determined by measuring the attenuation of UV light through a 50 cm long custom fabricated Beer Lambert cell. The ozone concentration can be varied from ppm to 25 1 ppm. [11] Initially an aqueous NaCl particle is captured by introducing an aqueous NaCl aerosol into the cell from an ultrasonic medical nebulizer. Once trapped the particle is held at constant RH (typically 75%) for typically 1 h. This period allows the RH to stabilize in the trapping cell, and the droplet size and composition to be determined from the CERS 2of13

3 spectrum. The trapped droplet is then doped through coagulation with OL aerosol generated by nucleation from an OL vapor formed above a heated reservoir of OL (Sigma- Aldrich, 90% purity) and introduced into the trapping cell in a flow of nitrogen. Collisions between OL aerosol and the trapped aqueous droplet can be observed by bright field imaging. After dosing, the trapped particle is visibly distorted in the bright field image and the Raman signatures of OL are visible in the CERS spectrum. The vapor pressure of OL is very low ( Pa [Cappa et al., 2008a]) and no significant evaporation of OL is observed on the time scale of the experiments (days). [12] The bright field image provides only limited information from the two-dimensional projection of the particle shape. The CERS spectra can provide complementary information on the composition, structure, and size of trapped droplets [Buajarern et al., 2007; Kwamena et al., 2010; Mitchem and Reid, 2008; Reid et al., 2007], and can be used to probe the structure of both effloresced and deliquesced NaCl/OL particles [Dennis-Smither et al., 2012]. Spectra are acquired by collecting the inelastically backscattered light from a trapped particle and imaged with a spectrograph/ CCD camera. Spectra are recorded with a 1 s acquisition time, but typically five spectra are summed before analysis, aiding the identification of weak spectral features. [13] CERS spectra are formed from a superposition of spontaneous and stimulated Raman scatter. The spontaneous Raman scatter, in the form of broad bands that are Stokes shifted from the incident radiation at 532 nm, can be used to probe the composition of the trapped particle by identification of vibrational modes characteristic of molecules within the particle. The stimulated Raman scatter arises due to the ability of the trapped particle to act as an optical cavity and leads to the formation of sharp peaks in the CERS spectra at wavelengths commensurate with whispering gallery modes (WGMs). In a homogeneous spherical particle, the wavelengths of the WGMs can be used to determine the size and refractive index (RI) of the particle by comparison of the observed WGM wavelengths with the wavelengths as predicted using Mie theory [Miles et al., 2012; Mitchem et al., 2006]. If the particle is distorted away from spherical shape, the particle is no longer a high-quality optical cavity and the intensity of the stimulated peaks is suppressed. For example, the loss of spherical symmetry in particles with a partially engulfed structure, such as deliquesced NaCl/OL/aqueous aerosol, leads to the suppression of scattering at WGM wavelengths [Buajarern et al., 2007; Dennis-Smither et al., 2012]. The light intensity associated with a WGM is concentrated within a shell at the surface of the particle and does not penetrate deeper than r-r/n from the particle surface, where r is the total particle radius and n is the real part of the RI of the particle [Eversole et al., 1995]. Therefore, for particles with a small inclusion at the center of the particle such as effloresced NaCl/OL particles the wavelengths of the WGMs are the same as those predicted for a homogeneous OL droplet. [14] Prior to doping with OL, the size of the aqueous NaCl droplet is determined at a reference RH allowing the mass loading of NaCl to be calculated from knowledge of the concentration at that RH, estimated from the Extended Aerosol Thermodynamics Model (E-AIM) [Clegg et al., 1998; Wexler and Clegg, 2002] (also S. L. Clegg et al., Extended AIM aerosol thermodynamics model, 2012, The mass of NaCl is reported as an equivalent dry particle volume for the inorganic seed, assuming that the density of the crystalline seed is the same as bulk NaCl (2.17 g cm 3 [Haynes, 2012]). The uncertainty in the dry NaCl volume is estimated to be up to 5%, mainly arising from uncertainty in the reference RH at which the droplet size is measured. The NaCl is assumed to be inert during the oxidative aging of the particles. [15] Figure 1 shows the CERS spectra of a deliquesced NaCl/ OL particle during oxidative aging, and the Raman spectrum of bulk OL for comparison. These spectra are collected using a 300 groove mm 1 grating. Bands corresponding to C-H and O-H stretching vibrations are visible between nm and nm, respectively, corresponding to Stokes shifts of cm 1 and cm 1.Peaks characteristic of OL can be identified at 583 nm (1655 cm 1 ) due to the carbon-carbon double bond stretching vibration [De Gelder et al., 2007] and at nm (3008 cm 1 ) due to the vinylic carbon-hydrogen bond stretching vibration [Lee and Chan, 2007]; that is, the stretching vibration of carbonhydrogen bonds on sp 2 -hybridized carbon atoms clearly indicates the incorporation of the OL into the trapped particle. An estimate of the relative amount of OL in the particle can be made from the intensities of the CH and OH bands in the CERS spectra, also allowing the changing composition to be followed as oxidative aging proceeds. Distortion of the mixed phase deliquesced particle away from spherical symmetry prevents determination of droplet size. [16] Also shown is the CERS spectrum of an effloresced NaCl/OL particle during oxidative aging. Unlike the deliquesced particles, which have a structure formed of two partially engulfing liquid phases, the effloresced particles have a spherically symmetric structure leading to the appearance of stimulated Raman peaks in the spectra. During the oxidation of effloresced particles a 1200 groove mm 1 grating was used. This covers a much smaller range of wavelengths with a wavelength resolution fine enough to accurately identify the position of WGMs allowing the particle size to be estimated with subnanometer resolution as oxidative aging proceeds. The characteristic OL peak at nm is observed in the region covered by the 1200 groove mm 1 grating. The integrated intensity of this peak above the baseline is normalized to unity before exposure of the particle to ozone, (I vch ), and is used to monitor the relative amount of OL remaining in the particle during oxidative aging. None of the products formed by the oxidative aging of OL are expected to have unsaturated carbon-carbon double bonds. 3. Results and Discussion [17] We first examine the rate of loss of OL during oxidative aging. Sections 3.1 and 3.2 discuss the time-dependent changes in the volume and RI of the organic phase, respectively. Section 3.4 presents observations of the change in particle size in the early stages of the exposure and reaction of OL with ozone. In section 3.5, we report measurements of the hygroscopicity of the organic phase before and after oxidative aging and the kinetics of water partitioning Kinetics of Loss of OL During Oxidative Aging [18] Experimental details for the measurements of oxidation of both deliquesced and effloresced NaCl/OL particles over 3of13

4 Figure 1. (a) CERS spectra of a deliquesced NaCl/OL particle during oxidative aging and the Raman spectrum of bulk OL for comparison. (b) CERS spectra of an effloresced NaCl/OL particle during oxidative aging. The dashed lines indicate the positions of features in the spontaneous Raman scatter characteristic of OL. ranges of both RH and ozone concentration are given in Table 1. All particles were exposed to ozone until I vch dropped to zero, consistent with the complete reaction of OL with ozone. This was confirmed for one particle, which was exposed to a further dose of ozone once all the volatile reaction products had evaporated after the first exposure. No change in the particle size was observed during the second period of exposure to ozone (1 h), with any radius change being <1 nm. [19] Figure 2 shows I vch for four of the effloresced NaCl/ OL particles during oxidative aging. Each data point gives the mean value of I vch over a 200 s time period, with the error bars indicating the standard deviation in the measurement. The OL Raman signal is clearly lost more rapidly with increasing ozone concentration. The error bars are larger for data points acquired during rapid loss of OL: reaction of OL is accompanied by a decrease in the particle size and the progression of successive WGMs across the vinylic C-H bond stretching band, increasing the error associated with the spontaneous band intensities. [20] For all the particles, the point at which all the OL has reacted and I vch reaches zero can be readily identified. Figure 3 shows the time taken from the start of the ozone exposure for I vch to reach zero as a function of ozone concentration for all the particles in Table 1. For both effloresced and deliquesced seeds, the time taken for complete loss of the OL Raman signal increases with decreasing ozone concentration. The time taken for the complete loss of the Raman signal will depend on the structure of the particle (affecting surface area-to-volume ratio) and the volume of the organic phase. Complete loss of the OL Raman signal occurs more quickly for deliquesced particles. In general, the reactions on deliquesced particles incorporated a smaller volume of OL and the partially engulfed structure leads to a higher surface area-to-volume ratio for the organic phase [Reid et al., 2011]. Thus, the OL is expected to react to completion on a shorter time scale when the inorganic seed is deliquesced. In the partially engulfed structure, the majority of the OL forms a discrete organic phase with a minor fraction forming a monolayer at the air/aqueous interface and acting as a surfactant [Reid et al., 2011]. This provides an additional surface area for the uptake of ozone and reaction of the OL. However, the uptake coefficient of ozone by an OL monolayer on an aqueous surface is 100 smaller than for pure OL particles [King et al., 2009] and the reactive loss of OL at the aqueous surface is not expect to be significant compared to the reaction in the organic phase. [21] To compare our results to previous studies, the kinetics of the reaction were interpreted using the resistor model treatment [Smith et al., 2002; Worsnop et al., 2002] in order to determine the reactive uptake coefficient for ozone on pure oleic acid, g OL. The majority of previous studies have concluded that the reaction is a near-surface reaction, limited by the diffusion of ozone into the condensed phase. Under this assumption, I vch is expected to obey a dependence on the ozone exposure time, t, of I vch ¼ 1 3g OLcP O3 t 2 ; ð1þ 8rRT½OLŠ 0 where c is the mean kinetic speed of O 3 molecules in the gas phase, P O3 is the partial pressure of O 3, r is the particle radius, R is the gas constant, T is the temperature and [OL] 0 is the initial concentration of OL in the particle [Smith et al., 2002]. Fits to this kinetic model are shown in Figure 2 for the four I vch decays shown, covering a broad 4of13

5 Table 1. The Dry Volumes of NaCl and OL in the Mixed NaCl/OL Particles and Experimental Conditions Used During Oxidative Aging Experiments Particle Initial Dry OL Volume (10 16 m 3 ) Particle Size Initial Dry NaCl Volume (10 17 m 3 ) [O 3 ] (ppm) RH (%) Experimental Conditions T (K) Total Length of Ozone Exposure (s) Particles Oxidized While Effloresced A < ,213 B < ,434 C < ,121 D < ,382 E < ,653 F < ,370 G < ,306 H < ,295 I < ,257 J < ,257 K < ,339 L < ,181 M < ,120 N < O ,255 P ,465 Q ,417 R ,653 Particles Oxidized While Deliquesced S ,604 T ,407 U ,870 V ,703 W ,463 X ,066 Y ,871 Z ,704 AA ,935 range of [O 3 ]. The best fit values of g OL are in the range ( ) 10 4 and have a mean value of This is marginally lower than the values for g OL reported from aerosol studies, which range from to (see Last et al. [2009] and Zahardis and Petrucci [2007] for compellations of g OL values from aerosol studies; we exclude the g OL of Broekhuizen et al. [2004] of as they state their value is likely an underestimation of g OL ). The full range of values reported in the literature span the range to [Last et al., 2009; Smith et al., 2002]. [22] As observed in Figure 2, it is expected that the kinetic fits deviate at long time from the experimental decays. The particles undergo significant size and compositional change during oxidation and equation (1) can only provide a first-order treatment [Worsnop et al., 2002]. We do not attempt to draw any conclusions from the form of the decay in I vch, only making the assumptions inherent to equation (1) so as to calculate g OL and allow convenient comparison with previous studies. Due to the complex interaction between changes in particle composition and size, a more explicit model of mass transport between gas and condensed phase is required, such as the KM-GAP model developed by Shiraiwa et al. [2012], to capture the full temporal evolution of both particle composition and size. [23] To compare the kinetics for loss of OL with previous measurements, we have calculated the halftime for the decay in I vch (t 1/2 ). The ozone exposure required for the reaction to proceed to the halfway point can be calculated by multiplying this halftime by P O3. In Figure 4 we report the dependence of this product on particle diameter for all of our measurements Figure 2. The decline in the normalized spontaneous Raman signal intensity of the OL vinylic C-H stretch (I vch ) versus time during oxidative aging by ozone under dry conditions (RH < 5%) and best fit curves for equation (1) for particles A (red, [O 3 ] = 0.6 ppm), C (green, [O 3 ]= 1.7 ppm), J (blue, [O 3 ] = 6.0 ppm), and N (brown, [O 3 ]= 20.8 ppm). Ozone exposure starts at 0 s. 5of13

6 Figure 3. The time taken from start of ozone exposure until complete loss of OL Raman signal during oxidative aging by ozone of effloresced (filled circles) and deliquesced (open circles) NaCl/OL particles as a function of ozone concentration over a range of RHs. with an effloresced seed and compare with a selection of previous studies for which we are able to estimate t 1/2 [Hearn et al., 2005; Katrib et al., 2005a; Last et al., 2009; Lee and Chan, 2007; Morris et al., 2002; Smith et al., 2002; Ziemann, 2005]. Also shown are lines indicating P O3 t 1/2 calculated using equation (1) for a number of different values of g OL. As stated previously, literature values for g OL lie in the range to Our results are broadly consistent with previous measurements, including the single particle measurements of Lee and Chan [2007] Changes in the Volume of the Organic Phase During and Post Oxidative Aging [24] In addition to measuring the rate of loss of OL, the radius of the organic host droplet can be determined using the wavelengths of WGMs in the CERS spectra from droplets with an effloresced NaCl seed [Mitchem and Reid, 2008]. Then, from the dry volume of NaCl calculated prior to dosing with OL, the volume of the organic phase during oxidation can be determined. Figure 5 compares the change in I vch and the organic volume for a single effloresced NaCl/ OL particle. As discussed earlier, the oxidative aging of OL leads to the formation of a variety of products with a range of volatilities and the evaporation of the highervolatility products leads to a decrease in the particle volume. [25] Associated with the change in composition during chemical aging and loss of volatile products, the RI of the organic phase will change. Our initial analysis of the change in particle size/volume over time will neglect this change in RI, an approximation that is satisfactory for most of our analyses. Indeed, we will show that particles with a radius of 4.5 mm decrease in radius by >1 mm. The change in refractive index during this time is <1%, equivalent to an error incurred in the final size estimate of 50 nm at most. More accurate fits of refractive index and size will be described later. [26] The loss of the OL Raman signal and the evaporation of volatile products clearly occur on two different time scales indicating that the partitioning of volatile products into the gas phase occurs more slowly than their rate of formation from OL ozonolysis. At 200 s, both the OL Raman signal and the volume of the organic phase have decreased with evaporation of some of the products from the ozonolysis within seconds of their formation, consistent with the formation of NL. I vch drops to zero after a few thousand seconds of ozone exposure whereas the decrease in the organic volume continues for several thousand seconds after this point. This slow change in the organic volume is consistent with the evaporation of NA and ON. Because there is a gas flow passing through the cell, volatilized compounds are removed from the trapping cell and, thus, the products do not form an equilibrium partitioning between the gaseous and condensed phases. After around s most of the volatile products have been lost from the particle and no significant changes Figure 4. Values of the product P O3 t 1/2 with varying particle diameter for the aging of NaCl/OL particles with an effloresced seed and for previous experimental studies. Lines show values of P O3 t 1/2 for fixed value of g OL. The particle diameters in the study by Lee and Chan [2007] are not measured, but they are estimated to be between 40 and 70 mm. Figure 5. The normalized spontaneous Raman signal intensity for the OL vinylic carbon-hydrogen stretch, I vch, (black) and the volume of the organic phase (red) versus time during oxidative aging by ozone of particle J under dry conditions. The shaded area indicates the period when the particle was exposed to ozone with ozone exposure starting at 0 s. 6of13

7 Figure 6. The time dependence of the organic volume (normalized to unity prior to ozone exposure) during oxidative aging by ozone of effloresced NaCl/OL particles under dry conditions (RH < 5%) over a range of ozone concentrations. Ozone exposure starts at 0 s. in the particle volume are observed; the organic volume is then formed of components that appear to be involatile. Typical changes in the particle volume occur at less than 0.1% per hour over a further s (8 h). [27] Figure 6 shows the time dependence of the normalized organic volume for the 14 effloresced particles oxidatively aged under dry conditions, calculated assuming a fixed value for the RI of the organic phase equal to that of OL (1.46 [Haynes, 2012]). As expected, the change in the normalized organic volume is more rapid for high ozone concentrations as the loss of OL, and hence formation of volatile products, occurs more quickly. [28] Previous studies have observed the formation of four major C 9 ozonolysis products (NL, NA, ON and AA) along with higher molecular weight products. The vapor pressures of NL, NA and AA have been determined experimentally as Pa (at K [Verevkin et al., 2003]), 0.2 Pa (at K [de Kruif et al., 1982]) and Pa (at 296 K [Bilde et al., 2003]), respectively. To our knowledge the vapor pressure of ON has not been experimental determined. By comparison of its structure to AA and NA we would expect it to have a volatility between that of AA and NA and indeed its vapor pressure can be estimated as Pa (at K) using group contribution methods [Nannoolal et al., 2004, 2008] implemented by E-AIM (S. L. Clegg et al., Extended AIM aerosol thermodynamics model, 2012, By consideration of the mass flux of each of these compounds during the quasi steady state evaporation [Pope et al., 2010] from a 6 mm radius particle, typical of the particles is this study, the complete evaporation of NL, NA and ON occur on time scales on the order of 10 0 s, 10 2 s and 10 3 s, respectively, whereas the evaporation of AA occurs on a time scale on the order of 10 7 s. Therefore, we would expect to observe evaporation of NL, NA, and ON. The loss of AA and other low vapor pressure/higher molecular weight products are expected to be much less than 0.1% per hour and their evaporation will not be noticeable during our experiments. [29] The branching of the reaction between volatile and involatile products can be compared as a function of ozone concentration. Although the trend is not unambiguous, the data in Figure 6 do suggest that a decreasing ozone concentration may lead to an increase in the involatile fraction. Slower ozonolysis kinetics may be consistent with an increased yield of higher molecular weight products through a reduction in the loss rate of organic components from the particle. [30] In order to estimate accurately the yields of involatile products formed from the changes in organic volume, we must consider changes in the organic phase RI and density. As discussed earlier, errors incurred from assuming that the RI of the organic phase is constant are small and will be addressed in section 3.3. However, we must also consider how the density of the organic phase changes during oxidative aging. Assuming that the density does not change significantly allows the decrease in the particle mass due to evaporation of volatile products to be deduced directly from the observed decrease in organic volume. An increase in the particle density through chemical reactions would also lead to a decrease in the particle volume. Katrib et al. [2005b] observed the density of the organic domain formed by the oxidative aging of OL layers on polystyrene cores to increase by 20%. Thus, the observed volume decrease of between 15 and 50% cannot be fully explained by a change in density and must be accompanied by volatilization of products. [31] Vesna et al. [2009] reported the percentage carbon yields for the four major C 9 ozonolysis products with the remaining carbon assigned as unidentified products (UPs), of which they attributed the majority to be higher molecular weight peroxides. To convert these carbon yields to percentages of the initial organic volume for comparison with our results, we assume volumes of the species are additive. Additionally, we require knowledge of the molar volume and number of carbons per molecule of OL, NL, NA, ON, AA and the UPs. Literature values are available for the molar volume of OL and the four major C 9 ozonolysis products. For the UPs, we assume a proxy with the molecular formula C 18 H 34 O 6. This is chosen as a value typical of the a-acyloxyalkyl hydroperoxides (AAHPs), C 9 dimer compounds that have been identified as products of the secondary reactions between the CIs and carboxylic acids [Ziemann, 2005]. The density of the proxy is assumed to be 1000 kg m 3. Vesna et al. [2009] determined carbon yields under dry conditions for NL, NA, ON, AA and UPs as 51.2, 2.8, 13.8, 6.9 and 25.3%, respectively. [32] Converting these carbon yields to percentages of the initial organic volume gives 55.8, 3.1, 14.8, 6.7 and 27.7%, respectively. The total is 108.2% indicating that if they remained in the condensed phase the products would occupy a volume 8.2% greater than the initial volume of OL. The involatile products (i.e., AA and UPs) are predicted to have a volume of only 34% of the initial organic volume compared with the 50 70% of the initial organic volume in our experiments. Assuming AA and the UPs form in the same ratio as measured by Vesna et al. [2009], our results would correspond to a carbon yield of AA + UPs of 47 65% compared to 32% measured by Vesna et al. [2009]. This suggests either a much greater branching to involatile products in our experiments or that the volatile products do not evaporate to the same extent on coarse mode particles when compared to the experiments on accumulation mode particles. Although 7of13

8 Figure 7. The time dependence of the RI (black circles) during oxidative aging by ozone of particle J. The radius determined using these RIs (red circles) and the radius determined using a fixed RI of 1.46 (red line) are also compared. The shaded region indicates the period when the particle was exposed to ozone. Vesna et al. [2009] used an ozone concentration of 0.5 ppm, less than used here, the particles sizes were much smaller (geometric mean diameter of 78 nm) and evaporation of products would be expected to be much more rapid. Hence, using the argument presented earlier, the rapid volatilization of products could reduce the formation of involatile higher molecular weight products. Bulk phase diffusion limited mass transport could also limit the evaporative flux and preserve a greater fraction of the organic mass in the particle for longer than for accumulation mode particles [Bones et al., 2012]. However, the kinetic limitation would be severe, far more than observed for water transport in our previous work [Bones et al., 2012; Tong et al., 2011]. We shall return to a discussion of this when we explore the kinetics of water transport during condensation. [33] Although the sample of particles studied is small and different RHs and ozone concentrations have not been studied extensively, there is no evidence of an influence of RH on the kinetics of OL loss during oxidation for particles with an effloresced NaCl seed (Figure 3), consistent with the observations by Vesna et al. [2009] Changes in the Refractive Index During Oxidative Aging [34] The organic volumes shown in Figure 6 were estimated by determining the total particle size assuming a fixed RI equal to that of OL. However, the wavelengths of the WGMs can be used to determine both the size and RI of the particle simultaneously, with no prior knowledge of either quantity [Miles et al., 2012]. In Figure 7 we report the time-dependent RI of the organic phase during the oxidative aging and evaporation of volatile products for a single effloresced NaCl/OL particle. We also compare the total particle radius determined using this best fit approach for RI and radius with that when assuming a fixed refractive index (1.46). The RI reported in Figure 7 is the mean of the best fit RIs determined from five consecutive CERS spectra (each the sum of five consecutive spectra acquired over 1 s), fitted without accounting for dispersion in the RI. The error bars indicate the standard deviation of the 5 values. [35] During the oxidative aging process there is an overall increase in the RI from to These results are similar to experiments by Cappa et al. [2011], who observed an increase in the RI of organic aerosol aged by OH. The exposure of the particles to ozone in our experiments is completed within seconds of turning off the ozone supply. However, the change in the calculated RI suggests that the particle continues to evolve for several tens of the thousands of seconds. Katrib et al. [2005b] observed the density of the organic domain formed by the oxidative aging of OL layers on polystyrene cores to increase with continued ozone exposure when OL is no longer present, indicating the condensed phase products of the reaction between OL and ozone may react with ozone and/or themselves. Additionally Broekhuizen et al. [2004] observed that the CCN properties of ozone-processed oleic acid particles continued to evolve during further ozone exposure. In our experiments, the RIs of oxidized particles continue to change without further ozone exposure and as discussed in section 3.1 exposure of the involatile organic volume to a further dose of ozone results in no significant change in the organic volume. [36] Comparison between the particle radii determined using a fixed value of the RI (1.46) and determined using the best fit RI shows that the radii agree well initially, but as the best fit RI increases further from its initial value the radii begin to diverge. The organic volumes calculated earlier, as shown in Figure 6, assumed a fixed value of the RI. An increase in the RI of during the change in the organic volume, as observed in Figure 7, suggests that the fractions of the organic volume remaining involatile are overestimated by up to 2.4%, a marginal error. [37] The RIs of NL, NA and AA are , and , respectively, whereas the RI of ON has not been measured. These RIs are lower than that of OL so the RI of the organic phase may be expected to decrease during oxidative aging. However, we have already shown that we would expect all of these products except AA to evaporate from the oxidatively aged particles on a time scale of <10000 s and the organic phase that remains to be composed of involatile products of which the RI is not known but may be higher than OL. [38] There are a number of possible explanations for the observed change of the RI over long time frames. First, there may be a continual change in the composition either through evaporation of or reactions between products leading to a change in the RI. However, evaporative loss or reaction would need to change the refractive index and density in such a way that the size remained approximately constant. An alternative explanation is a slow change in the homogeneity and internal structure of the particle. To retrieve the RI from the WGM wavelengths, it must be assumed that the particle is both spherical and homogeneous in composition [Miles et al., 2012]. However, if the light associated with the WGMs experiences a radial inhomogeneity in the RI within the organic phase that changes with time, because of changes in the structure or composition of the organic domain, then this would lead to a change in the apparent RI. [39] To consider further the possibility that the slow change in RI is due to a restructuring/change in homogeneity in particle composition, the characteristic half-time (t 1/2 ) 8of13

9 Figure 8. (a) The organic volume calculated using the wavelengths of the WGMs (filled circles) or the movement of one WGM (unfilled circles) for particle J upon initial exposure of the particle to ozone. Ozone exposure starts at 0 s. (b) The ozone concentration dependence of the increase in the organic volume at the beginning of ozone exposure, as a percentage of the initial organic volume. The oxidative aging by ozone is in each case for effloresced NaCl seeds under dry conditions. of mass transport and mixing by molecular diffusion in the particle can be estimated by ð t 1=2 ¼ r Particle r NaCl Þ 2 p 2 ; ð2þ D lnð2þ where r Particle is the radius of the particle, r NaCl is the radius of the spherical NaCl core and D is the binary diffusion coefficient. For the particle considered in Figure 7, t 1/2 for the change in RI is approximately s, r Particle = m, r NaCl = m, suggesting that a diffusion coefficient on the order of m 2 s 1 would be consistent with the observed relaxation in the particle RI. A diffusion coefficient of this magnitude corresponds to a highly viscous semisolid [Shiraiwa et al., 2011]. Given that the conditions are dry for the experiment reported in Figure 7 (RH < 5%), the organic phase (containing a high mass fraction of involatile oligomeric products) could indeed be highly viscous and the slow relaxation in RI and, thus, morphology, would be consistent with this Changes in the Particle Size at the Start of Oxidative Aging [40] Figure 8a reports apparent changes in the organic volume immediately after the ozone flow is introduced for the same particle shown in Figure 4. The organic volume has been calculated in two ways: from the particle size determined from the wavelengths of the WGMs in each spectrum and also by tracking the wavelength of a single WGM through the spectra and using this to determine the change in the particle size. There is a time delay (approximately 45 s) from the ozone being turned on at 0 s before a response in the particle size is observed, the time required for ozone to pass along the gas lines and enter the trapping cell. Surprisingly, the initial response of the particle is for the WGMs to move to higher wavelengths reaching a maximum at approximately 70 s, before then moving to lower wavelengths. This behavior is observed for all effloresced NaCl/OL particles exposed to ozone. It is assumed that the movement of WGMs to lower wavelengths is due to the decrease in particle size because of the partitioning of volatile reaction products to the gas phase. However, the initial movement of the WGMs to higher wavelengths must be attributed to either an increase in the organic phase volume, an increase in the particle RI or a combination of both. [41] For the particle reported in Figure 8a, the wavelengths of the WGMs increase by 0.14 nm. Using a fixed value of the RI this corresponds to an increase in the organic volume of 0.07%. Alternatively if the particle volume is assumed to be constant, an increase in the RI of 0.02% occurs. During the period when the WGMs move to higher wavelengths the scatter in the calculated RI is around 0.2%. Thus, it is not possible to determine whether the observed shift in WGMs is due to an increase in the particle volume, an increase in the particle RI or a combination of both. An increase in the organic volume could be caused by the absorption of ozone or the formation of products from the ozonolysis of OL. Any increase in the RI is probably due to the formation of products from the secondary chemistry as the absorption of ozone into the organic phase would be expected to lead to a decrease in the RI as would the formation of the four major C 9 ozonolysis products which all have lower RIs than OL. [42] Recent work using kinetic models for gas-particle interactions [Shiraiwa et al., 2010, 2012] has shown that an ozone concentration gradient is formed within the particle initially upon exposure of an oleic acid particle to ozone. During the first few seconds the ozone concentration in the near-surface bulk is up to 2 orders of magnitude higher than in the particle center [Shiraiwa et al., 2012]. The gradient in the ozone concentration gradient is expected to relax over a period of tens of seconds [Shiraiwa et al., 2010, 2012]. Therefore the reaction between ozone and OL and the formation of products is expected to occur near the particle surface and any change in the RI due to the formation of products would be expected to lead to a radial variation in the RI of the organic volume. [43] The equilibrium partitioning of ozone into the organic volume is not expected to lead to significant changes in the organic volume. However, as determined earlier, if all the products of the oxidative aging remained in the condensed 9of13

10 Figure 9. (a) The RI of the organic phase of an effloresced NaCl/OL particle at different relative humidities before (filled circles) and after (unfilled circles) oxidative aging. The lines show the RI calculate from the observed trend in the radial growth factors assuming a linear mixing rule for the RI of the mixed organic/aqueous phase before (solid line) and after (dashed line) oxidative aging. (b) The radius growth factor of the organic phase at the different relative humidities before (filled circles) and after (unfilled circles) oxidative aging. (c) The radius growth factor of the organic phase and the measured RH during a step change in RH before and after oxidative aging. phase the organic volume would be expected to increase by 8.2%. Therefore, the formation of products is expected to lead to a short-lived increase in the particle volume before the subsequent evaporation of the volatile products. [44] Figure 8b shows the maximum percentage initial increase in the organic volume upon first exposure of the effloresced NaCl/OL particles to ozone under dry conditions. The strong correlation between ozone concentration and magnitude of the volume increase suggests that the initial movement of WGMs to higher wavelengths is due to the initial formation of products occurring before significant evaporation has occurred Hygroscopicity of Organic Phase Before and After Oxidative Aging [45] We have previously probed the hygroscopicity of the organic fraction before and after oxidative aging of effloresced NaCl/OL particles by measuring the particle size at different RHs assuming a fixed value of the RI [Dennis- Smither et al., 2012]. We now present a reanalysis of some of these data based on our ability to extract both the particle size and the RI of effloresced mixed NaCl/OL particles. Figure 9 shows the RH dependence of the RI of a single effloresced NaCl/OL particle before and after oxidative aging (and subsequent evaporation of volatile products). Each reported RI is the mean and standard deviation of the best fit RI for 20 consecutive CERS spectra, each the sum of 20 consecutive spectra acquired over 1 s. The best fit RI is determined assuming the dispersion in the RI with wavelength shows a linear dependence over the range of interest, 589 to 660 nm [Millard and Seaver, 1990]. The RH is measured by the probe situated immediately prior to the trapping cell and the horizontal errors bars represent an uncertainty of 2%. The RI of the organic phase prior to oxidative aging is , slightly higher than the value for the RI of OL used earlier of [46] Figure 9 also compares the mean values with standard deviations of the radius growth factor of the organic phase before and after oxidative aging. The organic phase is clearly more hygroscopic after oxidative aging. We can use the observed trends to calculate the expected change in the RI of the organic phase with increasing RH. The lines in Figure 9a show the predicted RI trend assuming the volume of the liquid phase is the sum of the volumes of the organic component and water and the RI can be approximated by a linear mixing rule. The decrease in RI is in good agreement with the observed RI, both before and after oxidative aging. This suggests that water uptake forms a mixed organic/aqueous phase that is homogeneous in composition. [47] Our previous reported radius growth factors for the organic phase in effloresced particles before and after oxidative aging did not account for the decrease in the RI with increasing RH [Dennis-Smither et al., 2012]. Assuming a fixed value of the RI, the largest radius growth factor we previously calculated was Now, accounting for the decrease in the RI due to the uptake of water for this growth factor, assuming linear mixing rule for the RI, we can estimate the actual growth factor was [48] The initial time response of the effloresced NaCl/OL particle to a step change in the RH of the gas flow passing through the trapping cell before and after oxidative aging is shown in Figure 9c. Both before and after oxidative aging, the particle responds on a time scale similar to that of the RH probe, suggesting that the particle takes up water rapidly and there is no evidence of a kinetic limitation to the water 10 of 13