The evolution of surface structure during simulated atmospheric ageing of nano-scale coatings of an organic surfactant aerosol proxy †

Atmospheric aerosol particles can be coated with organic materials, impacting aerosol atmospheric lifetime and urban air quality. Coatings of organic materials are also found on indoor surfaces such as window glass. Oleic acid is a fatty acid surfactant that is abundant in cooking and marine aerosol emissions. Under ambient conditions it can self-assemble into lamellar bilayers (stacks) with its sodium salt. We found that nano-scale oleic acid – sodium oleate ﬁ lms spin-coated onto solid silicon substrates form a mixed-phase area of lamellar stacks and amorphous ﬁ lms. The coatings were subjected to simulated atmospheric ageing (ozonolysis and humidity changes) while the surface structure was followed by neutron re ﬂ ectometry. We found that the orientation of lamellar stacks, which is known to a ﬀ ect the di ﬀ usivity of small molecules through them, was sensitive to humidity both in oxidised and pristine ﬁ lms. Lamellar bilayer stacks in oxidised ﬁ lms acquired (cid:1) 11-fold more water under humid conditions (>80% relative humidity) compared to the unoxidised ﬁ lm, demonstrating a signi ﬁ cant increase in ﬁ lm hygroscopicity after oxidation. Lamellar stacks, consisting only of starting materials, persisted at the end of simulated atmospheric ageing. These ﬁ ndings for atmospherically relevant nano-scale ﬁ lms corroborate previous work on micrometre-scale layers, thus demonstrating that fatty acid self-assembly could signi ﬁ cantly increase the atmospheric lifetime of these molecules. The persistence of such semi-solid surfactant arrangements in the atmosphere has implications for the climate as well as urban and indoor air pollution.


Introduction
Aerosols contribute to ambient air pollution and affect processes such as cloud droplet formation, affecting air quality and climate. 1,2Organic compounds can dominate aerosol composition in many parts of the world. 35][6][7] Poor air quality has been linked with an increased organic fraction on the surface of urban particulate matter 8 and cooking emissions have been estimated to contribute 10% to PM 2.5 emissions in the UK. 9 Away from the urban environment, organic compounds have been characterised in particulate matter from biogenic sources such as remote marine 10 and forested 11 environments.The study of the effect of organic compounds on aerosol processes is therefore of global importance.
Organic surface coatings on particulate matter have been characterised in eld measurements. 10,12The reactivity of organic molecules with OH radicals has been shown to proceed faster as a particle surface coating compared with pure particles, linked with the higher surface area-to-volume ratio associated with such coatings. 13The chemical lifetime of lms of insoluble atmospheric organic materials at the air-water interface has been predicted to range from minutes to days with respect to the OH radical. 14This could have an impact on the residence times of particulate matter, as organic coatings have been shown to prolong the lifetime of pollutants 15 and affect particle water uptake. 16These observations highlight the importance of understanding how a particle surface lm structure responds to atmospheric processing, which we address in this study.
Some organic emissions are surface active, such as fatty acids.The unsaturated fatty acid oleic acid is commonly found in cooking and marine aerosols [17][18][19] and fatty acids have been identied on the surface of particulate matter. 10,20,212][33] Oleic acid molecules, being amphiphilic, can arrange into a range of lyotropic liquid crystal (LLC) phases in contact with water and in the presence of its sodium salt. 34,35They can also form anhydrous lamellar bilayers (stacks) at laboratory relative humidity (RH) ($50% RH). 23These molecular arrangements bring with them differences in viscosity and diffusivity, two factors which impact atmospheric aerosol ageing. 36,37ndoor surfaces can host organic compounds which are emitted by common indoor activities such as cooking, cleaning and ironing. 38,39Organic lms were shown to form on glass surfaces aer exposure to cooking emissions and qualitatively shown to include oxidised and unoxidised fatty acids. 40It has been suggested that lms collected in a kitchen change surface morphology and/or form viscous phases during humidity changes. 41Both hypotheses (morphology change and viscous phase formation) have implications for the uptake of atmospheric oxidants such as ozone due to the implied changes in surface area and lm diffusivity.Oleic acid is likely present in such coatings because oleic acid and its reaction products have been followed in real-time during and aer simulated cooking 18 and it is also used as a tracer for cooking emissions. 4The surface coatings presented here can therefore be a proxy for the surface of atmospheric aerosols and organic coatings on indoor surfaces.
4][45][46][47] We have shown previously that the lamellar bilayer stack formed by the oleic acid-sodium oleate proxy slows down ozonolysis by ca. an order of magnitude, with thicker lms reacting slower. 23The orientation of such bilayer stacks is expected to affect the diffusion of small molecules through them.Diffusion perpendicular to the bilayer plane is orders of magnitude smaller than in-plane diffusion. 48Bilayers are found in the dehydrated upper layer of the skin (stratum corneum) along with dead cells.In the context of indoor air quality, models of skin lipid-oxidant reactivity consider diffusion through this layer due to the marked decrease in molecular diffusivity, 49 which can be partially attributed to the lipid bilayers present. 50,51These anisotropic arrangements can therefore have an impact on molecular diffusivity through them.The orientation of such bilayer arrangements is followed in this study.
3][54][55][56][57][58][59] These surface-sensitive techniques allow for the study of organic lms down to the monolayer level.NR can be particularly sensitive to a sample if it is deuterated, due to the much larger neutron scattering power of a deuterium atom compared with a hydrogen atom.
Previous work has focussed on monolayers of deuterated oleic acid on an aqueous sub-phase, 27,55,59 self-assembly in an oleic acid-sodium oleate proxy in large levitated particles 24,60 and reaction kinetics in capillaries coated with micron-scale lms of this lamellar phase proxy down to ca. 0.6 mm. 23inetic modelling of these experiments has shown that the chemical lifetime of oleic acid could increase by days upon selforganisation. 30Many of these studies have suggested the persistence of reacted and/or unreacted organic materials aer chemical ageingsomething we observe and discuss here.
In this study, we spin-coated nano-scale lms of deuterated oleic acid-sodium oleate mixtures on silicon blocks with a native SiO 2 layer.This resulted in a coating with a mixed area composed of lamellar stacks and an amorphous region.We subjected these lms to simulated atmospheric ageing by oxidising the sample and changing the humidity.Changes in the surface structure were followed by NR and complemented by grazing-incidence small-angle X-ray scattering (GI-SAXS) and optical microscopy.We highlight the atmospheric implications derived from these results.Our system and methodology are designed to act as a platform for the future, and more quantitative studies of the reaction kinetics and surface dynamics of such coatings.

Methods
Most of the techniques presented here are not commonly used in atmospherically relevant studies.The key points necessary for the understanding of each technique and what they probe are included here.More details are included in the relevant sections of the ESI.† 2.1 Preparation of spin-coated lms Deuterated oleic acid (d 34 -OA, Sigma-Aldrich 98% Atom D; 99.9% purity) was used in order to provide a large contrast between the reected neutron signal from the deposited lm and the substrate required for the neutron reectometry experiment.Sodium oleate (Sigma-Aldrich, 99% purity) was not deuterated.Methanol (Sigma-Aldrich, 99.8% purity) was used as the solvent for sample solutions.5 cm polished silicon disks (Crystran, UK) were used as the substrates for these experiments.Full details of the sample preparation and spin-coating procedure are in the ESI (Section S1 ESI †).The substrate surfaces were characterised before lms were deposited on them by NR and XRR, a summary of which is in the ESI (Section S2 ESI †).In total, 5 lms were coated at various spin speeds (1 Â 1000 rpm, 2 Â 2000 rpm and 2 Â 4000 rpm).

Neutron reectometry (NR)
NR is described in detail elsewhere. 61,62Here, we summarise the salient points of the technique.NR was carried out at the ISIS neutron and muon source (UK) on the INTER beamline and on the FIGARO beamline at the Institute Laue-Langevin (France).A schematic of the NR experiment is presented in Fig. 1(a)-note that the x-z plane is the specular plane.
NR is a technique used to probe the structure of interfaces by providing depth-resolved structural information in the form of a scattering length density (SLD) prole.Information on the thickness, roughness and density of each interfacial layer can be obtained by NR.Different isotopes have different neutron scattering lengths.The SLD of a layer is a function of the scattering length and volume fraction of each chemical component in that layer.Layered structures, e.g. the oleic acid-sodium oleate lms on silicon studied here, give rise to steps in SLD.
Fig. 1 Schematic representations of (a) neutron reflectometry (NR) and (b) grazing-incidence small-angle X-ray scattering (GI-SAXS) experiments.The GI-SAXS data presented are from a film coated on a silicon wafer at 2000 rpm.The mixed area model is illustrated, showing regions of the amorphous film and lamellar bilayers (stacks).The relationship between lamellar stack orientation and scattering pattern is illustrated in (b).The X-rays and neutrons travel along the x-axis in the positive direction, the x-z plane is the specular plane, and the angle of incidence (q) is identified in panel (a).
The reectivity (Rthe fraction of neutrons reected) is related to the SLD and momentum transfer (Q) via the formula: This relationship is valid signicantly above the critical angle, below which all neutrons are reected.Eqn (1) is the high-Q limit for a single layer of thickness d in air.In the case of solid-supported layers, as is the case in this study, eqn (1) is still valid if the SLD of the substrate is twice the SLD of the layer, which is roughly the case here.In general, however, the specular reectivity has to be solved analytically using well-known optical matrix formalisms as performed here. 63For specular reectometry, Q is related to the neutron wavelength (l) and angle of incidence/reection (q): Details of the NR experiments carried out in this study are presented in the ESI (Section S1, ESI †).
The chamber used to control the sample environment during NR experiments on INTER is described by Skoda et al. 58 In summary, the chamber: has a volume of approximately 1.5 L; has quartz windows at either end, which allow the neutrons to pass through, hit the sample and be detected on the other side; the chamber has in-and outlets, which allow the chamber to be purged with a gas-such as ozone and water vapour used in this study.The chamber used on FIGARO followed a similar design with a slightly smaller volume of 1 L and sapphire windows along the neutron path.

Grazing-incidence small-angle X-ray scattering (GI-SAXS)
GI-SAXS is closely related to NR in that small (grazing) angles of incidence are utilised to probe the surface structure of materials. 64The GI-SAXS experiment allows for the measurement of X-ray scattering over a larger angular range and also measures off-specular scatteringthis is scattering in the y-direction in Fig. 1(b).This allows us to determine the orientation of the lamellar stacks (see Fig. 1(b)).GI-SAXS was carried out on the I22 beamline at the Diamond Light Source, UK. 65 Details of the GI-SAXS experimental method are in the ESI (Section S1, ESI †).

Ozonolysis and humidity experiments
Ozone was generated by passing dry oxygen through a precalibrated UV pen-ray ozoniser (Ultraviolet Products Ltd, Cambridge, UK) and then to the sample chamber in which the sample was placed.
Humidity control was achieved either using a bespoke Raspberry Pi (RPi)-based system (experiments on INTER) or by varying pre-calibrated air ows of wet (D 2 O) and dry N 2 ows (experiments on FIGARO).Further details of the ozonolysis and humidity control experiments are in the ESI (Section S1, ESI †).

NR model tting
NR data are commonly analysed by tting interfacial models to the experimental data. 61,62Fitting rather than direct inversion is required due to the loss of phase information in the scattering process.
For all ts performed in this study, a mixed area model was selected, where the area illuminated by the neutron beam has regions of different interfacial structures (see Fig. 1 for a schematic representation).This model, consisting of a lamellar stack region and an amorphous lm region, t the best to all the lms probed in this study.This was backed up by determination of the Bayesian evidence for each model, [66][67][68][69] complementary optical microscopy and GI-SAXS (see succeeding discussion and Section S1, S5 and S6, ESI †).
The refnx Python package 63 was used to create and t the model to the data, optimising the model parameters with a global optimisation genetic algorithm. 70Markov Chain Monte Carlo (MCMC) sampling of the parameter space was employed via the emcee Python package. 71Full details of the interfacial model, model tting procedure, MCMC and Bayesian evidence determination are provided in the ESI (Section S5, ESI †).

Results and discussion
3.1 The surface structure of nano-scale lms Each spin-coated lm was characterised before being subjected to simulated ageing.The resulting NR curve and SLD prole for measurements taken at low and high angles (stitched together) are presented in Fig. 2 for a lm coated at 4000 rpm.Note that the lamellar stack region accounts for 7% of the observed signal (see Table 1).

Paper
Environmental Science: Atmospheres The R vs. Q curve presented in Fig. 2(a) is typical of a layered interface.Fringes appear at regular intervals and the spacing between them is inversely proportional to the lm thickness. 61A Bragg peak at ca. 0.14 ÅÀ1 is observed and is consistent with the anhydrous lamellar phase Bragg peak observed in a previous SAXS study on the same non-deuterated proxy, corresponding to a lamellar bilayer thickness of ca.4.5 nm. 23n order to quantify our condence in our proposed model, Bayesian evidence (Z) estimation for a series of plausible model interfacial structures (including the mixed area model) was performed using the nested sampler available in the dynesty Python package in combination with refnx. 63,69We found the greatest evidence for the mixed-area model (Fig. S5, ESI †).Details of this analysis are in the ESI (Section S5, ESI †).The mixed area model is further corroborated by the analysis of spin-coated samples prepared away from the beamline on $3 cm 2 silicon wafers.Optical microscopy shows a mixed area with islands in the order of $50-100 mm in diameter and complementary GI-SAXS measurements of those lms exhibit lamellar stacks with both parallel (to the substrate surface) and random orientations (see Section S6, ESI †).This agrees with NR and synchrotron GI-SAXS measurements (see later).
The model reectivity curve and SLD prole are generated using optimised parameters (Fig. 2).Additionally, curves derived from the MCMC sampling procedure that are consistent with the data illustrate the uncertainty associated with the optimised parameters and model ts.Other initial ts are presented in the ESI (Section S3, ESI †).Variations between samples in the initial amorphous lm SLD are likely due to a varying ratio of deuterated oleic acid to non-deuterated sodium oleate present in the amorphous lm.
The initial proportion of the sample area occupied by lamellar stacks oriented parallel to the surface was found to be between 1 and 15% for the lms studied here, and the rest of the sample area being the amorphous lm (Table 1).This was derived from the scale factors applied to each component of the mixed area model.Note that this does not mean that only 1 to 15% of the lm is self-organised into lamellar stacks.These are tted to specular NR curves: only reected neutrons from repeating lamellar stacks that are parallel to the surface are detected.Model ts to experiments carried out on FIGARO returned thicker lms with lower proportions of parallel lamellar stack area.The relationship between spinning speed and lm thickness persists (i.e.lower spinning speeds result in a thicker lm).However, we cannot rule out the effect of the coating environment (temperature and humidity) and substrate on the initial lm structure.
Characterisation of the same system (with non-deuterated oleic acid) by GI-SAXS revealed a diffuse scattering ring in addition to the specular lamellar stack peak (see Fig. 1(b)).We are therefore condent that the lamellar stack region of the lm is a mixture of both parallel and randomly oriented lamellar stacks.We discuss changes in lamellar stack orientation during simulated ageing in Section 3.2.

Simulated atmospheric ageing of nano-scale lms
We simulated atmospheric ageing by exposing the spin-coated lms to ozone (126-4020 ppb) and humidity (5-90% RH).Film thickness, roughness and the relative amount (scale factor) of each component can be followed using the NR model tting procedure.All the results are summarised in Fig. 3.The NR prole of the lm coated at 4000 rpm was monitored under a dry oxygen ow before oxidation to rule out potential surface structure changes only due to oxygen (Fig. 3(c), (f), (i), (l) and (o)).
3.2.1 Ozonolysis.Parallel oriented lamellar stacks persist during and aer extensive oxidation (see the Bragg peak at $0.14 ÅÀ1 in Fig. 3(a)-(c)).We corroborated this nding with full high and low angle NR measurements on a lm aer oxidation, showing a Bragg peak (Fig. S7, ESI †).The Bragg peak decreased in intensity during ozonolysis (Fig. 3(a)-(c)), which is consistent with observations made during ozonolysis of micronscale lms of the non-deuterated proxy. 23The implications for the persistence of such surfactant organic materials are discussed in Section 4.
A slight decrease in lm thickness is observed during the initial stages of ozonolysis, most evident for the lm coated at 2000 rpm (Fig. 3(d)-(f)).The ozonolysis of oleic acid yields nonanal as one of the primary products.Nonanal is volatile and is assumed to be lost from the lm aer formation, accounting for the decrease in lm thickness. 28Generally, the drop in lm thickness is small relative to the initial lm thickness.a This is the number of repeating lamellar stacks which are oriented parallel to the substrateonly information on the parallel oriented lamellar stacks is available from these specular NR measurements.This number is also the nearest whole number of repeating stacks tted due to the MCMC sampling returning the mean number of parallel lamellar stacks consistent with the data.b Data collected on the FIGARO beamline at the ILL, France.The change in lm thickness when increasing the ozone concentration is small and negligible for the lm coated at 2000 rpm (Fig. 3(e)).As ozone exposure continues, non-volatile reaction products are likely to remain in the lm.Assuming the small decrease in lm thickness is mostly due to the loss of nonanal from the lm, the initial ozonolysis reaction is either complete or the formation of inert reaction products hinders further oleic acid ozonolysis.Note that this would be valid for the amorphous oleic acid lm region and not the lamellar phase oleic acid region, which is by denition unreacted.3][74][75] The formation of such crusts has been postulated previously and has implications for the ageing of aerosol particles. 76,77 change in lm thickness is observed for the lamellar phase region during ozonolysis (Fig. 3(m)-(o)).Generally, there is a decrease in lamellar region thickness as ozonolysis proceeds and this is clearest for the lms coated at 1000 and 2000 rpm (Fig. 3(m) and (n)).The trend is less clear for the lm coated at 4000 rpm (Fig. 3(o)).This region is modelled as stacks of lipid bilayers, and therefore any signal coming from this structure is assumed to be from the oleic acid-sodium oleate bilayer as they are the only molecules in abundance able to form a bilayer.This assumption is valid because if the 9-carbon products from the reaction form bilayers, the associated Bragg peak would appear at a much higher Q (shorter d-spacing).This has not been observed here or in SAXS experiments during ozonolysis of the same non-deuterated system. 23he lm roughness increases with exposure to ozone (Fig. 3(g)-(i)).This is consistent for all lms except for the lm coated at 4000 rpm and aligns with changes in lm morphology observed during the ozonolysis of pure oleic acid deposited on a substrate. 78This implies that the surface area associated with these lms also changes with time and that the surface morphology changes somewhat during ozonolysis.The atmospheric signicance of this change in surface roughness is discussed in Section 4.
The relative amount of the amorphous lm and oriented lamellar stacks changes during ozonolysis.The scale factors applied to each region were constrained so that their sum was the original total scale factor before ageing (AE0.02)determined by measuring the critical edge (see Fig. 2(a) maximum R region at low Q).There is an apparent decrease in the amount of oriented lamellar stacks relative to the amorphous lms coated at 2000 rpm (Fig. 3(k)), though this effect is less pronounced for the thicker lm coated at 1000 rpm (Fig. 3(j)).We reiterate that this scale factor is not a measure of how much of the area is covered by lamellar stacks, it only concerns the area covered by parallel oriented lamellar stacks.This is of interest due to the anisotropic diffusion characteristic of molecules through such lamellar bilayers. 48.2.2Humidication and dehumidication of oxidised and unoxidised lms.More changes in the surface structure occur when humidifying and dehumidifying the oxidised lms.The surface structure changed rapidly and substantially with the Bragg peak shiing to a lower Q (higher lamellar d-spacing) (Fig. 3(a)-(c)).This is due to the lamellar stacks taking up water, resulting in a thickening of the lamellar bilayers (Fig. 3(m)-(o)).Dehumidication reversed this trend and the Bragg peak moved back to its original position.This movement of the Bragg peak was reversible, suggesting that the lamellar stacks can take up water readily and reversibly (Fig. 3(a)).Regarding amorphous lm thickness, the trend is not entirely reversible (Fig. 3(d)).
Reorganisation of the lm and possible de-wetting from the surface is evidenced by the loss of the critical edge during humidity changes (Fig. S7 and S9, ESI †).The loss of the critical edge during humidication suggests that in-plane correlations in the surface structure occur due to surface de-wetting.This results from off-specular scattering around the critical edge, decreasing the observed specular NR signal. 79An increase in offspecular scattering around the critical edge has been demonstrated for a polymer lm de-wetting from a surface. 80We carried out off-specular NR measurements, showing evidence of off-specular scattering occurring around the critical edge, which increased at high humidity (see Section S9, ESI †).
Reversible orientation of the lamellar stacks was observed for both oxidised and unoxidized lms of the same composition.This is evident from the increase in the overall scaling factor arising from the lamellar stack region during humidication (Fig. 3(j)-(l)).Off-specular NR measurements revealed a strong specular Bragg peak signal at 82% RH for both oxidised and unoxidised lms (Fig. 4(c) and (d)).This strong signal was not present at lower RH.There is also the appearance of a Bragg sheet at high RHa diagonal streak of intensity across the specular Bragg peak (Fig. 4(c) and (d)).This is evidence of a degree of random orientation of the lamellar stacks.
Controlled humidity experiments suggest that this reversible Bragg peak shi occurs at $80% RH (Fig. 4(c) and (d)).Measurements at RH values below this did not reveal any signicant change in the size and position of the Bragg peak and Bragg sheets are not visible in the off-specular measurements at low humidity (<35% RH) for both oxidised and unoxidised lms (Fig. 4(a) and (b)).This is also the case for an intermediate humidity (Fig. S8, ESI †measurement at 63% RH).Complementary GI-SAXS measurements conrm the random orientation of lamellar stacks at room relative humidity (57% RH) (Fig. 5).
The oxidised lm lamellar bilayers take up more water than the unoxidised lm at high RH.The position of their respective lamellar Bragg peaks shows this (Fig. 4(e)lower Q corresponds to a higher d-spacing, see eqn (S1) †).From the Bragg peak position, the lamellar phase d-spacing at 82% RH is 44 Å and 63 Å for the unoxidised and oxidised lms, respectively.Assuming that the oleic acid-sodium oleate bilayer thickness is constant (42 AE 2 Å measured from NR model tting of initially dry spincoated lms), the thickness of the water layer at this humidity is 2 and 21 Å for the unoxidised and oxidised lms, respectively.This corresponds to bilayer stacks which take up $11-fold more water when oxidised.Oleic acid lms oxidised by ozone can be signicantly more hygroscopic than unoxidised oleic acid lms 81 and more oxidised compounds, measured using the O : C ratio, are known to be more hygroscopic. 82We suggest that oxidised reaction products may have dissolved into the aqueous layer between lamellar bilayers, resulting in a more hygroscopic mixture and therefore a thicker aqueous layer.However, it was not possible to determine this directly with our NR experiments.
Complementary GI-SAXS measurements on the nondeuterated and unoxidised forms of this proxy at high RH show lamellar stack orientation and that an inverse micellar phase forms and exhibits a diffuse scattering peak at a higher Q ($0.2 ÅÀ1 ).Lamellar stack orientation was followed by plotting scattered intensity vs. the azimuthal angle in the Q-range of the rst order Bragg peak (see Section S9 and Fig. S9, ESI †).The ratio between specular and diffuse scattered intensity (I specular / I diffuse ) is higher at high RH (an increase of $30%).This effect is visible in the 2-D GI-SAXS patterns (Fig. 5).This higher ratio indicates an increased proportion of lamellar stacks parallel to the substrate plane, corroborating the NR results.Note that the underlying reected X-ray beam in the specular direction also contributes to I specular , and therefore this is a qualitative measurement.
The inverse micellar phase observed would have different physical properties compared with the lamellar stacks, 34 adding to the range of factors which change while ageing these organic coatings.Note that this inverse micellar phase peak would not be distinguishable by NR measurement due to the limited Qrange of NR experiments (see eqn (1)).

Atmospheric implications
Cooking emissions evolve over time and consist of both fresh and oxidised organic materials (including oleic acid). 18Organic coatings on indoor glass surfaces can contain a mixture of oxidised and non-oxidised compounds. 404][45][46] In addition to the effect of lamellar phase formation on reactivity, 23 we have shown here that lamellar stacks found in both fresh and oxidised lms of our proxy can orient themselves upon humidication.The diffusion of small molecules in the lamellar phase is directionally dependent, with diffusion parallel to the bilayer plane orders of magnitude higher than in the perpendicular direction. 48,83This suggests that lamellar stacks oriented parallel to the surface present an extra hindrance to the diffusion of small molecules, such as ozone, through the organic lm.This humidity-dependent change in oleic acid orientation could therefore affect the atmospheric ageing of lms and particles containing oleic acid.While we did not specically study the effect of surface hydration on the morphology of the deposited organic lm, hydrated surfaces are abundant in indoor air and hydrated and non-hydrated surfaces are likely to exhibit different behaviours.A hydrated surface is more hydrophilic, and therefore the wetting behaviour of hydrophobic organic lms over such surfaces could be affected by surface hydration.
The oxidation of organic coatings proceeds rapidly compared to the bulk reaction of the same species as pure particles. 13Here, the increase in lm roughness observed during oxidation implies that there is an increase in organic surface area.This larger surface area may facilitate faster uptake of reactive gases such as ozone, though information on reaction kinetics could not be extracted from these NR experiments.Deposited droplets of oleic acid have previously been observed to change morphology during ozone exposure 78 and oating monolayers of oleic acid-stearic acid can restructure during ozonolysis. 58Such changes in organic lm morphology are not widely considered in models of aerosol particle and lm reaction kinetics, 25,26 though the very recent development of a kinetic multi-layer model of lm formation, growth and chemistry (KM-FILM) has demonstrated a shi towards such a consideration. 84Data from reectometry measurements may help constrain such a model in the future.
An organic coating consisting of unreacted self-assembled oleic acid and reaction products remains deposited aer simulated atmospheric ageing.This is consistent with a recent study on oating monolayers of oleic acid, where surface active products remained at the air-water interface aer exposure to ozone, 27 demonstrating that this persistence of the organic lm occurs both at the air-water and air-solid interfaces.The residual lm is likely to be a mixture of oxidised reaction products. 28The lamellar stack Bragg peak also remains, con-rming that lamellar phase oleic acid withstands prolonged exposure to ozone 23 and that this effect is now observed in nanoscale lms on a solid surface.5,86 We provide direct experimental evidence that lamellar bilayers are more hygroscopic aer lm oxidation.The link between the oxidation state and hygroscopicity is well-established in aerosol science. 82,87Here, we are able to estimate the effect of oxidation on the nanostructure formed by this aerosol proxy at high humidity (>80% RH).At this humidity, $11-fold more water is present in the bilayer stacks aer lm oxidation compared to a "fresh" unoxidised mixture.This apparent increase in hygroscopicity suggests that water uptake and eventual cloud droplet nucleation are more likely to occur for these mixtures when oxidised.Oleic acid ozonolysis can result in the formation of highmolecular-weight compounds. 72,74The product lm that remains aer ageing may contain these oligomers, resulting in a more viscous lm.An increase in viscosity has been observed experimentally for this reaction system and has been linked with the formation of such oligomers. 88A viscous organic material in the surface layers of particulate matter has been linked to the increased long-range transport and persistence of harmful compounds found in urban particulate matter, increasing the risk to human health. 15,89This protection is due to the reduced diffusivity of atmospheric oxidants such as ozone, limiting the extent of particle ageing.Together with our previously determined effect of self-assembly on reactivity in mm lms, 23 the persistence of unreacted semi-solid materials (lamellar oleic acid) following simulated ageing in this study demonstrates that this effect is also valid for atmospherically more relevant nano-scale lms.

Conclusions
This study has demonstrated that surface coatings of a semi-solid organic aerosol proxy undergo changes in morphology during simulated atmospheric ageing.These changes are expected to affect the uptake of trace gases into the condensed phase, especially on the nano-scale due to the increase in the surface area-tovolume ratio caused by the increasing lm roughness.We highlight the importance of considering lm morphology in model and experimental studies on nanoscale lms.
Lamellar stacks formed by the fatty acid reversibly oriented themselves parallel to the surface with increasing humidity.This was found for both oxidised and non-oxidised lms, with oxidised lms taking up $11-fold more water at high humidity compared with non-oxidised lms.
The lamellar stacks persisted aer simulated atmospheric ageing.This is consistent with previous ndings on mm-scale lms.Additionally, an amorphous lm which likely consists of oxidation products also remainsconsistent with recent work on oating oleic acid monolayers and conrming that lm retention is also occurring at air-solid interfaces.These experiments thus demonstrate that the persistence of the semi-solid lamellar phase is valid over the micro-to-nanoscale and that oxidation products can persist down to the monolayer scale, depending on conditions.
Crucially, the NR technique does not require molecules to be assembled into 3-D nanostructures or aggregates, so that future experiments on more complex proxies and real atmospheric materials deposited on solid surfaces are possible, as long as suitable coatings can be made that are stable for the extended measurement times required for non-deuterated samples.

Fig. 2
Fig. 2 (a) R vs. Q curve of a film coated at 4000 rpm.A model fit is presented along with 200 curves randomly sampled from the chains stored after the MCMC sampling procedure (black curvesvery close to the model fit line).(b) SLD profile corresponding to the amorphous film and lamellar stack regions.SLD is plotted vs. the distance from the substrate surface (z).The 200 model fits from the MCMC sampling procedure are also presented to illustrate the uncertainty (fainter lines).Note that the lamellar stack region accounts for 7% of the observed signal (see Table1).

Fig. 3
Fig. 3 Plots of RQ 4 vs.Q curves measured during simulated atmospheric ageing of the 3 films studied here: 1000 rpm coating (a), 2000 rpm coating (b) and 4000 rpm coating (c).Optimised parameters from model fits are plotted for each experiment: film thickness (d)-(f), film roughness (g)-(i), fraction of parallel lamellar stacks (j)-(l) and lamellar region thickness (m)-(o).The neutron beam went down for a while during film oxidation of the 4000 rpm coating (c)parameters generally were unchanged after this period.The coloured regions correspond to ozonolysis at 126 ppb & 4020 ppb, humidification (Hum.), dehumidification (Dehum.)and dry O 2 .

Fig. 4
Fig. 4 Off-specular NR measurements on separate unoxidised ((a) & (c)) and oxidised ((b) & (d)) films at low and high humidity (see top right of each panel for exact RH).The specular direction is denoted by the dashed red line and the specular Bragg peak is highlighted by red circles in panels (c) and (d).(e) A comparison of 1-D specular NR curves for the oxidised and unoxidised films at 82% RH.A schematic of the lamellar bilayer is also presented along with the d-spacings derived from the Bragg peak position.

Fig. 5
Fig. 5 2-D GI-SAXS patterns of an unoxidised film of the nondeuterated analogue of this proxy at 57% RH (a) and 84% RH (b).The scattering intensities arising from parallel (Para.)and random (Rand.)lamellar orientations are labelled accordingly in the 57% RH pattern.The inverse micellar scattering ring (Mic.)present at 84% RH is also labelled for clarity.The 2-D GI-SAXS patterns are corrected for the incident X-ray intensity and are mapped on the same intensity scale.

Table 1
Parameters obtained from initial model fits to NR data from spin-coated samples