Introduction

Atmospheric aerosols, solid or liquid particles suspended in air, can be primary, from direct emissions, or secondary, formed through gas-to-particle conversion of low-volatility vapours. Fine particles with a size of less than 1 mm account for 50-70% of the total particulate matter mass under highly polluted environments. 1 In particular, ultrane particles (diameter D p < 100 nm) are suspected to pose a substantial health risk for humans due to their smaller size and thus higher body penetration 2 but are difficult to chemically characterize due to their low mass. A substantial fraction of the atmospheric particle number concentrations are created by new particle formation (NPF), which is thought to be a source for up to half of the global cloud condensation nuclei (CCN). 3,4 The growth of nucleated particles to the CCN size is dominated by the condensation of low-volatility oxidation products from biogenic or anthropogenic precursors. Secondary organic aerosol (SOA) is a highly complex matrix of thousands of individual chemical species, the majority of which are present only in trace amounts. 5 Moreover, rapid intra-particle reactions which increase the complexity of SOA and the need for their real-time measurement have been proposed and observed. 6,7 While a variety of chemical ionization mass spectrometry (CIMS) techniques have been able to provide real time measurement of gas phase composition, no comparable techniques were available for real time particle-phase measurements until recently. For example, CIMS-based techniques used for atmospherically relevant mass concentrations (detection limits of ng m À3 ), such as the Filter Inlet for Gases and Aerosols (FIGAERO), 8 Thermal Desorption Chemical Ionization mass spectrometer (TDCIMS), 9 or Thermal Desorption Differential Mobility Analyzer (TD-DMA) 10 require separate collection and analysis stages and are thus "quasi" real time. However, considering the sample collection time (>600 s), the time resolution of such techniques is a limitation for the identication of species responsible for the early stages of particle growth, as well as the detection of rapid intra-particle processes. Since they are based on thermal desorption, for these techniques thermal decomposition can also play a role, biasing measurements and complicating molecular identication. Nevertheless, they may provide additional physiochemical information such as aerosol volatility or size-resolved aerosol composition. Instruments like the Aerosol Mass Spectrometer (AMS) trade off the ability to be quantitative for extensive thermal and ionization-induced fragmentation due to high vaporisation temperature (600 C) and the use of electron ionization (70 eV). 11 The AMS cut-off size of $60 nm also hinders its ability to characterize ultrane particles. Additionally, the CHARON-PTR ("chemical analysis of aerosol online" inlet coupled to a proton transfer reaction timeof-ight mass spectrometer) 12 and the AeroFAPA-MS (aerosol owing atmospheric-pressure aerglow mass spectrometer) 13,14 have been developed to provide online particle-phase measurements without the need for separate collection and analysis stages. Although these instruments have sufficiently low detection limits for atmospheric measurements, they are limited by cut-off size drawbacks (>100 nm) and ionization induced fragmentation for the CHARON-PTR and competing ionization pathways for the AeroFAPA-MS, leading to complicated spectral interpretation.

In response to these limitations, the extractive electrospray ionization time-of-ight mass spectrometer (EESI-TOF) 15 has been recently developed, providing online molecular-level chemical identication of OA at detection limits of several ng m À3 , with minimal thermal and ionization-induced fragmentation. The EESI technique has been further advanced to allow for online tandem mass spectrometry (coupled to an Orbitrap mass analyser) and the characterization of water-soluble metals. 16,17 The real-time capability of the EESI-TOF has enabled measurement of gas-to-particle partitioning in chamber studies and provided evidence of condensed-phase reactions. 7,18 In this work, we used the EESI-TOF to study in real time the molecular composition of freshly nucleated particles from the oxidation of naphthalene or b-caryophyllene in the Cosmic Leaving OUtdoor Droplets (CLOUD) chamber at the European Organization for Nuclear Research (CERN). We determined the size and mass detection limits of the EESI-TOF and present a detailed intercomparison of the chemical composition measured with the EESI-TOF and FIGAERO-CIMS. Furthermore, we show that the measured particle composition is largely consistent with the condensation of the measured organic vapours as described by an aerosol growth model. Overall, we demonstrate that the EESI-TOF is well suited to study the composition of nanoparticles during their early growth.

Methods

CLOUD chamber

The measurements were carried out in the CLOUD chamber at CERN during the CLOUD11 campaign in fall 2016. The CLOUD chamber is a 26.1 m 3 electropolished stainless steel chamber which enables experiments at atmospheric conditions with a very low background contamination. A schematic of the CLOUD chamber is shown in Fig. S1, † and it is described in detail by Kirkby et al. (2011) 19 as well as Duplissy et al. (2016). 20 Ozone (O 3 ) is produced by owing a small air ow past a quartz tube surrounded by UVC lamps (wavelength < 240 nm). A UV light system, containing four 200 W Hamamatsu Hg-Xe lamps (wavelength 250-450 nm, adjustable power) and a 4 W KrF excimer UV laser (wavelength 248 nm), generates hydroxyl (OH) radicals via O 3 photolysis. To photolyse NO 2 to NO, strong LEDs were used (wavelength 385 nm). In order to avoid contamination, pure air was generated by the evaporation of cryogenic liquid nitrogen (Messer, 99.999%) and liquid oxygen (Messer, 99.999%) at a ratio of 79 : 21. This makes it possible to investigate aerosol nucleation and growth in a nearly contaminationfree environment. 21 Relative humidity was controlled by owing a fraction of the air through a Naon® humidier using ultrapure water (18 MU cm, Millipore Corporation). Two Teonvaned fans were mounted to the oor and ceiling of the chamber to ensure quick gas-phase mixing. The CLOUD chamber was operated in continuous ow mode. During typical experiments, a large suite of instrumentation is connected to the chamber to characterize gas-and/or particle-phase compounds spanning over wide volatility and/or size ranges. Gas-phase monitors measure the gas phase concentrations of O 3 (Thermo Environmental Instruments TEI 49C), SO 2 (Thermo Fisher Scientic Inc. 42i-TLE), NO (ECO 485 Physics, CLD 780TR) and NO 2 (CAPS NO2, Aerodyne Research Inc.). A commercial nano SMPS (TSI 3938) with a water CPC (TSI 3788) measures the dry aerosol size distribution from 4.6 nm to 60 nm. A home-built scanning mobility particle sizer (SMPS) with a differential mobility analyser (DMA) and a condensation particle counter (CPC, TSI 3010) measures the aerosol size distribution from 20 nm to 400 nm.

EESI-TOF

The EESI-TOF setup used in these measurements is based on the setup of Lopez-Hilker et al. (2019). 15 A backing pressure of 400 mbar was applied to the electrospray (ES) solution bottle, providing a ow of ES solution through a fused silica capillary with an inner diameter of 75 mm (BGB Analytik AG, Boeckten, Switzerland). The EESI-TOF inlet temperature was kept constant at 250 C throughout the experiments. A potential difference relative to the mass spectrometer interface in the range of 2.5-3 kV was applied to the ES solution to provide a stable ES signal. The ES working solution used was methanol/water (50/50 v/v), doped with 100 ppm NaI. This ensures that the analyte ions are detected predominantly as sodiated adducts ([M + Na] + ) in the positive ionization mode. Methanol/water ES solution may lead to an increased background signal (more detected ions from impurities) compared to more conventionally used acetonitrile (ACN)/water ES solution. 15 However, it is better suited for the detection of organic nitrates compared to ACN/ water solvents, where N-containing ions may correspond to ACN + Na clusters. Previous studies on infusion ESI report reactions between methanol solvent and carbonyl analytes, e.g. yielding acetals/hemiacetals, which are shown to worsen with sample storage time (minutes-days). 22 These reactions are not expected to be signicant in our EESI system due to the much shorter extraction timescale (milliseconds). The HTOF mass analyzer of the EESI-TOF had a mass resolution of $4000-4500 and was operated with 1 s time resolution with periodic lter blank measurements for 2 minutes every 13 minutes. EESI-TOF data were averaged to 10 seconds and background-corrected by subtracting lter blank measurements. As some of the tted peaks in the EESI-TOF data may correspond to background contaminations from the inlet or the sampling lines, a data lter was applied to identify genuine analyte ion signals: time series of all EESI-TOF identied ions were correlated against total SMPS mass, and ions with an R 2 value of less than 0.7 were excluded. Each molecular formula was also checked manually to make sure that no ions that could correspond to genuine SOA peaks were excluded, and only time series belonging to contamination peaks were eliminated.

The EESI-TOF signal was converted to mass ux reaching the detector (in ag s À1 ), by scaling with the molecular weight of each molecule i, MW i , as follows:

where EESI (Hz) is the signal directly measured by the TOF and N a is Avogadro's number. The detection efficiency in the EESI-TOF is dependent on a variety of factors including mass transmission to the detector, extraction efficiency, particle size and ionization efficiency. As we do not yet have a complete understanding of the effect of these factors, we uniformly apply a bulk sensitivity factor to convert the EESI-TOF signals to mass concentrations. Since this sensitivity will depend on the status of the mass spectrometer, it is obtained by calibrating the total EESI mass ux (ag s À1 ) with the bulk aerosol mass from the SMPS for each experiment.

The reported EESI signals are also normalized by the main reagent ion signal (NaINa + ) to account for uctuations in the electrospray stability during the different experimental runs.

FIGAERO-CIMS

The EESI-TOF measurements were compared to the particlephase measurements of a FIGAERO-CIMS, which measured both the gas-and particle-phase in alternating stages. During the gas-phase measurement stage, a separate sample stream with a ow rate of 6 L min À1 passed through a 24 mm diameter PTFE lter for particle collection for $50 min. To initiate the particle desorption stage, the PTFE lter was transferred to the measurement line and 2.7 L min À1 of pure N 2 were heated progressively to thermally desorb and vaporise the collected particles. The pure N 2 and thus the lter temperature increased from 20 to 150 C at a rate of $10 C min À1 . The desorption stage lasted for 14 minutes. In either stage, the gas-phase or desorbed analyte vapours were sampled into a 150 mbar ion-molecule reactor and ionized by iodide (I À ) ions generated by passing an N 2 ow containing CH 3 I, supplied using an inline permeation source, over a 210 Po radioactive source. An LTOF mass analyser with a mass resolution of approximately 10 000 was used for ion separation and detection. The organic analytes were detected predominantly in the form of iodide adducts [M + I] À (>95% relative abundance).

Gas-phase mass spectrometers

A proton transfer reaction mass spectrometer (PTR3-TOF-MS), described in detail by Breitenlechner et al. (2017), 23 used proton-transfer or ligand-switching to measure the precursor concentrations as well as oxygenated volatile organic compounds (OVOCs). Highly oxygenated molecules (HOM) and sulphuric acid (H 2 SO 4 ) were measured using a nitrate chemical ionization atmospheric pressure interface time-of-ight mass spectrometer (Nitrate-CI-APi-TOF), described in detail by Kürten et al. (2014), 24 using nitrate anions ((HNO 3 ) n (NO 3 À ), with n ¼ 0-2) as reagent ions.

Estimated volatility parametrizations

The saturation vapour concentration of organic compounds from naphthalene oxidation, C * was estimated using the parametrization described in detail by Wang et al. 25 based on FIGAERO-CIMS measurements,

where C * 300 is the saturation vapour concentration at 300 K in mg m À3 ; n C is the number of carbon atoms of the molecule, n 0 C ¼ 25 based on the reference carbon number of pure hydrocarbons with C * ¼ 1 mg m À3 . On average, each carbon addition decreases log 10 C * 300 by b C ¼ 0.48. Additionally, functional groups decrease log 10 C * 300 and are parameterized in the formula by the effective oxygen number (n O ). A nitrate group (-ONO 2 ) normally reduces log 10 C * 300 by 2.5, so for simplicity the nitrate group was counted as -OH. Therefore, the effective oxygen number n O was calculated as the oxygen number subtracted by twice the nitrogen number. For SOA originated from naphthalene oxidation, b O ¼ 1.72 is applied based on FIGAERO measurements. 25

Aerosol growth model

We modelled particle growth based on the measured gas-phase concentrations of the oxidation products and their estimated volatilities, as described by Xiao et al. 26 When combining the gas phase concentrations from the two mass spectrometers, signals from the PTR3 were used for monomers with a low degree of oxygenation (n C # 10, n O # 5) whereas signals from the NO 3 -CIMS were used for all compounds with n O $ 6. Signals from analyte clusters with (NO 3 ) À and HNO 3 (NO 3 ) À were summed up when obtaining the concentrations from the NO 3 -CIMS. Compounds measured in the gas phase by PTR3 and NO 3 -CIMS were grouped into volatility bins according to the VBS framework. 27 Using the measured gas-phase concentrations grouped into volatility bins, the particle-phase is modelled similar to the works of Stolzenburg et al. (2018) and Tröstl et al. (2016) as follows. 28,29 The driving force of gas to particle partitioning of each VBS bin is described as:

where C g,i is the measured gas-phase concentration of the i th VBS bin and C eq,i is the equilibrium concentration of the i th VBS bin. Mass or volume growth was modelled for aerosol particles with diameters above 6 nm, as the mass of particles below 6 nm is negligible. So, for simplicity we exclude the Kelvin term (which accounts for the curvature effect of very small particles).

Hence, C eq;i ¼ c i C * i , where c i is the activity. Assuming that the particle phase is an ideal solution, it follows that c i ¼ C p,i /C p,tot , which is the ratio of the particle mass in the i th bin (C p,i ) to the total particle mass (C p,tot ). Sulphuric acid was allowed to condense at the kinetic limit in the growth model.

The gas to particle condensation ux can be described as:

As we only model growth for aerosols with a diameter above 6 nm and thus ignore the Kelvin effect, the condensation rate K i is approximated by the condensation sink of the aerosol population:

where Therefore, the volume growth of aerosol can be described as:

where V is the total particle volume; k dil is the chamber dilution rate (k dil ¼ 1.6 Â 10 À4 s À1 ) and k wall is the particle wall loss rate.

We obtained the growth of aerosol particles in volume by solving the equations describing the gas-to-particle partitioning process.

Results and discussion

As the CLOUD project focuses mainly on new particle formation and growth, the mass concentrations in the chamber are generally low (<10 ng m À3 ). Therefore, the EESI-TOF mostly operated very close to its detection limits. 15 A series of experiments, including b-caryophyllene ozonolysis and naphthalene oxidation, were dedicated for particle-phase measurement by extending the growth period to achieve nal aerosol concentrations of 400-800 ng m À3 . Experimental conditions are given in Table 1.

Proof-of-principle measurements of naphthalene and bcaryophyllene SOA at CLOUD

To evaluate the suitability of the EESI-TOF to measure the composition of nanoparticles during early growth, we analysed the EESI-TOF response for a series of experiments of naphthalene oxidation by OH radicals, in the presence of NO x , a system characteristic of polluted urban areas. 32 Fig. 1a shows the particle size distribution during a typical nucleation and growth experiment at CLOUD (experiment 1). The experiment shown in Fig. 1a includes particles initiated from 2 growth events; it is not possible to separate the contribution from each event, since the EESI-TOF does not provide size resolved data. Nevertheless, the View Article Online mean volume-weighted particle diameter, as calculated by the SMPS, ranges between 20-60 nm throughout the entire experiment.

During this experiment, naphthalene oxidation product monomers (n C ¼ 8, 9, 10, red) and dimers (n C ¼ 20, blue) are observed by the EESI-TOF, shown in Fig. 1b, as the SOA mass concentration increases to $800 ng m À3 . As shown in Fig. 1c, while EESI-TOF and FIGAERO-CIMS show good agreement in the general trend of naphthalene oxidation products, as exem-plied by C 10 H 10 O 5,7 , the EESI-TOF is able to provide a more time-resolved description. In total, for this experiment we identied 125 ions (96 monomers and 29 dimers) out of which 33 were nitrogen-containing (27 monomers and 6 dimers) using the EESI-TOF, as well as 168 ions (120 monomers and 48 dimers) out of which 63 nitrogen-containing (47 monomers and 21 dimers) using the FIGAERO-CIMS. For experiment 1, 33 identical ions were detected by the EESI-TOF and FIGAERO-CIMS. The correlation of the trend of these common ions is given in Fig. S2. † Overall, the intensity of the common ions accounts for 46% of the total EESI-TOF intensity and 36% of the total FIGAERO-CIMS intensity (at maximum aerosol mass). A full mass spectrum of the identied ions, for 5 minutes at the period of maximum aerosol mass from experiment 1, is given in Fig. 1d. The particle-phase composition as measured by the EESI-TOF is consistent with the compounds found in the gas phase from the Nitrate-CIMS (Fig. S3 †), as well as previous work on gas-phase naphthalene oxidation products. 33,34 Additionally, the corresponding online measurements of the molecular composition of freshly nucleated SOA from naphthalene and b-caryophyllene, both in the absence of NO x (experiments 2 and 3) are shown in Fig. S4 and S5, † respectively. We observe excellent correlations (R 2 > 0.93) between the mass detected by the EESI and the bulk SOA mass determined from the SMPS for all experiments (Fig. S6 †). The calibration factor varies between experiments as it is dependent on both instrumental settings as well the SOA composition.

The intercomparison of EESI-TOF and FIGAERO-CIMS in Fig. 2 shows signicant overlap, adding condence to the molecular speciation of the organic aerosol responsible for nucleation and early growth of nanoparticles. Both instruments indicate that C 9-10 compounds with 8 or less oxygen atoms dominate the particle phase spectra. In the EESI-TOF, C 11 compounds only account for 1-2% of the total, suggesting that reactions between methanol in the electrospray solution and C 10 compounds (yielding acetals/hemiacetals from carbonyls) 22 are negligible. This is also consistent with the good agreement between EESI-TOF and FIGAERO-CIMS in terms of carbon number distribution in the monomer region, where the FIGAERO-CIMS is not affected by such reactions. We also observe good agreement in the oxygen distribution measured by the EESI-TOF and FIGAERO-CIMS, particularly for C 10 compounds. In addition, both instruments observe smaller compounds (C 5-8 ). Many of these, e.g. C 5-8 H 6-14 O 3-9 , are also present in the gas-phase as measured by PTR3 and/or nitrate-CIMS. It is possible that these are authentic naphthalene oxidation products or fragmentation artefacts, given their View Article Online relatively high volatility. Whereas the FIGAERO-CIMS (equipped with an LTOF) reported C 16-20 dimers, we preferentially assigned the EESI-TOF peaks in the dimer region (m/z > 300) to C 20 compounds, given the lower mass resolution of the HTOF mass analyzer coupled to the EESI. Overall, the signal corresponding to dimers is higher in the EESI-TOF than the FIGAERO-CIMS (13% of the total particlephase signal, Fig. 2c, compared to 2% for the FIGAERO-CIMS, Fig. 2d). This could be due to several reasons. First, it is possible that the FIGAERO-CIMS suffers from thermal decomposition fragmentation 28,35,36 of these (extremely) low-volatility compounds due to the higher temperature needed to desorb them from the lter. This possibility would agree with the observation that less oxygenated compounds (O 1 -O 3 ) are more abundant in the FIGAERO-CIMS spectra, which could correspond to the fragments as they are unlikely to condense. Here we have not performed a thermogram analysis to identify and correct for any possible thermal decomposition artefacts. Moreover, [2M + Na] + clusters could also form during ionization in the EESI-TOF, elevating the observed dimer fraction. As these clusters are typically minor (0.1-1% of [M + Na] + ) and difficult to distinguish from authentic dimers without the use of a declustering scan, we did not attempt to correct for this potential artefact. Finally, this inconsistency could also arise from other instrumental differences such as differing compound-specic sensitivity, mass transmission, different ionization mechanisms, and more.

Another discrepancy in the SOA composition measured by the two instruments is the lower fraction of nitrogen-containing species observed in the EESI-TOF (16% compared to 26% in the FIGAERO-CIMS). Also, despite the lower dimer fraction measured by the FIGAERO-CIMS, a higher number of nitrogencontaining dimers (24) was detected, compared to 6 in the EESI-TOF. Many of these nitrogen-containing species have low intensities and we assume that a large fraction of these species fall below the detection limits of the EESI-TOF. Further, it has been shown that hydroxynitrates can lose HNO 3 during the ionization process in the EESI-TOF, 18 leading to their classication as non-nitrogenated compounds, though it is not clear how nitrogen containing functional groups would behave in aromatic systems. Regardless, dimers containing nitrogen contribute less than 1% to the total aerosol signal in both instruments and as such do not affect the main ndings from the studied system. Fig. S7 † shows the chemical composition for naphthalene SOA without the addition of NO x (experiment 2), from both the EESI-TOF and the FIGAERO-CIMS. As expected, nitrogencontaining species are almost completely absent in the experiment without NO x , with only minor peaks detected, most likely due to the high background level of those ions from previous experiments. Additionally, NO x also appear to suppress dimers in the particle phase: a considerably smaller dimer fraction is measured in the presence of NO x by both the EESI-TOF (14% with NO x , 21% without) and FIGAERO-CIMS (2% with NO x , 4% without). This is consistent with previous studies showing a reduction of the dimer fraction by NO x in the gas-phase and consequently in the particle phase for both biogenic and anthropogenic SOA. 25,37 For the b-caryophyllene system (Fig. S8 †), the two instruments also show a good agreement, with C15 compounds making up the bulk of the detected ions. Similar to the naphthalene system, smaller compounds are also detected in both the EESI-TOF and the FIGAERO-CIMS. Both instruments agree that the dimer fraction is lower for the b-caryophyllene system as compared to the naphthalene system. It is possible that the relative dimer fraction will be smaller for b-caryophyllene as monomers are able to partition to the particle phase to a greater extent. This is because of the higher carbon number for b-caryophyllene (C 15 ) compared to naphthalene (C 10 ) and thus lower volatility (despite lower O : C ratio). It is also possible that RO 2 radicals with less functionalization (lower O : C ratio) of b-caryophyllene could have lower accretion reaction rates than the more functionalized (higher O : C) RO 2 from naphthalene. This observation would agree with the previous study of Berndt et al. (2017). 38 Quantitatively, a good agreement is also observed between the intensities of the ion signals (at maximum aerosol mass) in the EESI-TOF and the FIGAERO-CIMS. For the compounds that were detected by both instruments in experiment 1, this relationship is shown in Fig. S9. † While the sensitivity towards different compounds in both instruments is expected to vary slightly, the respective detected intensities for any species are always within an order of magnitude, suggesting good general agreement. Overall, there appears to be considerable overlap in the chemical composition of the SOA measured by both instruments for the SOA investigated here.

EESI-TOF size-resolved detection limits

The 2 minute limit-of-detection of the EESI-TOF (LOD) was calculated as three times the standard deviation of background measurements, while universal sensitivity was applied for all ions in each SOA system. This is an estimate and disregards iondependant sensitivities. The universal sensitivity was calculated by calibrating the total EESI mass ux (ag s À1 ) with the bulk aerosol mass from the SMPS (using a density of 1.4 g cm À3 ) for experiments 1 and 3, resulting in sensitivities of 0.0017 and 0.0047 ag s À1 ng À1 m 3 for naphthalene and b-caryophyllene SOA respectively (Fig. S6 †). A two-minute time period when the air passed via the particle lter of the EESI-TOF was considered as background measurement for the calculation of the LODs.

Estimated detection limits for the main ions observed by the EESI-TOF in the naphthalene and b-caryophyllene SOA systems are shown in Fig. 3a, on an ion-by-ion basis as a function of m/z. The LOD for all detected ions is less than 10 ng m À3 , with similar detection limits for naphthalene and b-caryophyllene SOA. Lower m/z compounds appear to have higher detection limits than larger, less volatile compounds such as dimers. As the signals determined by the EESI-TOF are calculated by subtracting the periodic lter blanks, the ion-specic background during these lter blank periods is crucial for accurate signal determination. During the lter measurement period, semivolatile compounds adsorbed to the inlet walls may repartition back to the gas phase, causing an elevated background compared to the low-volatility compounds, which would result in a lower signal reported than expected. Another possibility for the increased background in the lter period is gas breakthrough in the denuder. Although the charcoal denuder is >99% efficient in removing the gas phase, 15 the EESI-TOF has a much higher sensitivity for the gas-phase than the particlephase, 39 meaning that even a relatively small amount of gas breakthrough could elevate the background of the lter period. Some ions show a small step change in the background levels as gas-phase concentrations increased, suggesting that at least some ions are affected by gas-phase breakthrough. However, as particle growth follows immediately aer the increase in gasphase concentrations, it is difficult to differentiate between the effects of vapor-wall interactions and denuder breakthrough.

We characterize the detection limits of the EESI-TOF as a function of particle size in order to assess its ability to study freshly nucleated particles (Fig. 3b). The EESI-TOF followed the evolution of the particle size distribution during experiment 1, therefore detecting different particle sizes at different times. The signal-to-noise ratio (SNR) was calculated as the ratio of the signal to the noise as follows; similar to the LOD calculation, a two-minute lter blank period during the experiment was chosen as the background level. The signal was dened as a continuous time series of the background-corrected ion intensity observed by the EESI-TOF while the noise was dened as:

where s is the standard deviation and n is the number of data points. The majority of detected ions (119 ions, 97% of total ion signal at maximum mass) are above SNR ¼ 3 for particle diameters as small as 30 nm, corresponding to a total OA mass of $100 ng m À3 . Since just a few species dominate the particle phase signal, we are also able to reliably detect the key chemical species contributing to particle growth for particle diameters as small as 14 nm. For example, 81 ions, corresponding to 80% of total ion signal at maximum mass, are detectable above SNR ¼ 3 at such particle size and mass. These results show that the EESI-TOF can accurately characterize the chemical composition of ultrane particles in real-time, with sufficiently low detection limits on the order of ng m À3 to fully characterize the chemical composition at aerosol loadings of a few hundreds of ng m À3 .

Study of aerosol growth using the EESI-TOF for naphthalene SOA

We used a VBS-based approach to model aerosol growth based on the gas-phase concentrations of sulphuric acid and organic condensable vapours measured with the PTR-MS and the Nitrate-CIMS. Fig. 4 (and Fig. S10 †) shows the modelled particle mass as well as the measured particle-phase concentrations from the EESI-TOF, grouped according to estimated volatility: extremely low volatility organic compounds (ELVOCs), low volatility organic compounds (LVOCs), semi-volatile organic compounds (SVOCs) and intermediate volatility organic compounds (IVOCs).

As in our previous study, 26 we observe an excellent agreement between the modelled particle mass and the actual particle mass as measured by the SMPS (Fig. S10 †). In Fig. 4, we compare the observed and predicted organic mass fractions. Modelled particle-phase sulphate mass is subtracted from the measured particle mass concentration (by the SMPS) for comparison with the EESI-TOF, since condensed-phase sulphate species undergo Na + exchange during the EESI-TOF ionization process, forming [Na 2 SO 4 + Na] + instead of simple Na adducts (e.g. [H 2 SO 4 + Na] + ), and was therefore not included in the quantitative chemical composition analysis. We also obtain a good overall agreement between the predicted particlephase composition from the condensation of gas-phase species and the measured particle-phase composition from the EESI-TOF. For comparison, EESI-TOF signals were converted to estimated mass by calibrating the mass ux reaching the detector of the EESI-TOF against the total organic mass, therefore assuming uniform sensitivity for all measured components.

Since the aerosol growth model is strongly dependent on the estimated volatilities of the gas-phase oxidation products, we explored a range of parametrizations from previous studies (Fig. S11 †). 25,27,28 The original Donahue et al. (2011) 27 VBS parametrization, developed before the role of autoxidation in SOA was appropriately appreciated, attributes oxygen atoms to ]O and -OH functional groups, which decreases the saturation vapor pressure more than -OOH. This leads to an underestimation of volatility when autoxidation predominates the organic oxidation mechanism, and therefore results in an overestimation of the ELVOC and LVOC fractions. Stolzenburg et al. (2018) 28 adapted the parametrization to include the increased fraction of hydroperoxide (-OOH) products from autoxidation as well as covalently-bound dimers, both commonly found in a-pinene SOA. Wang et al. (2020) 25 used direct FIGAERO measurements of volatility to demonstrate that both parametrizations are valid, depending on the relevant chemistry: aromatics such as naphthalene lie close to the original parametrization, whereas biogenics such as terpenes lie close to the "autoxidation" parametrization. However, in Fig. 4 (a) Stacked modelled particle mass (above 6 nm) of total, ELVOC, LVOC, SVOC particle-phase naphthalene oxidation products. The ELVOC bin is an overflow bin and thus contains all compounds with log C * < À4.5. Total organic mass, calculated as measured SMPS mass minus modelled sulphate mass, overlaid as solid black line. (b) Stacked measured concentrations of total, ELVOC, LVOC, SVOC, IVOC particle-phase naphthalene oxidation products from the EESI-TOF, assuming uniform response factor. The ELVOC bin is an overflow bin and thus contains all compounds with log C * < 4.5. Total organic mass, calculated as measured SMPS mass minus modelled sulphate mass, overlaid as solid black line. is based on FIGAERO volatility measurements for this particular naphthalene SOA system, so this is the most relevant parametrization for this case and is used for the growth model unless specied otherwise. Fig. 5a shows a quantitative comparison of the different volatility classes from the growth model and measured by the EESI-TOF. The modelled results show that initial growth (before 08:15 AM) is dominated by ELVOCs (Fig. 5b). As some of the detected compounds are still under the detection limit of the EESI-TOF at the lowest mass concentrations, the EESI-TOF data is only shown for mass concentrations greater than $50 ng m À3 .

Both the modelled and measured results consistently show that the contribution from more volatile compounds such as LVOCs and SVOCs becomes more signicant as particle mass increases (Fig. 5b), with LVOCs being the most abundant compound class for this experiment. A good agreement between the modelled and measured LVOC and SVOC fractions is seen, however, ELVOCs seem to be more abundant in the EESI-TOF data than we would expect from the condensation of the gas-phase. One explanation for the discrepancy of the ELVOC fraction could be due to the assumption of uniform sensitivity in the EESI-TOF when converting signals to estimated mass concentrations, which may be too simplistic. The EESI-TOF sensitivity has been shown 31 to increase with increasing molecular weight and oxygen content (e.g. n C n O /(n C + n O )), and therefore inversely correlate with log C * (i.e. higher sensitivity for lower volatility compounds). Finally, we cannot rule out the possibility of condensed-phase reactions taking place which have not been considered in the aerosol growth model. The SOA composition has previously been observed to change aer condensation; 6,7 in particular, accretion reactions (e.g. C 7-10 + C 7-10 ¼> C 14-20 ) 40 would lead to an increase in the ELVOC fraction observed by the EESI which had not been taken into account in the model. However, due to the relatively short timescale of the experiment as well as the fact that all of the particle-phase compound concentrations appear to increase simultaneously in the EESI-TOF as the gas phase oxidation product concentrations increased, we are not able to distinguish such condensed-phase processes here. Interestingly, in the measured EESI-TOF data we also observe a small IVOC fraction that is not predicted by the growth model, which may be caused by gas-phase compounds breaking through the denuderas the particle mass increases the relative contribution of this fraction decreases as the gasphase source is rather constant. We note that although the denuder is highly efficient at removing the gas phase (>99%), 15 even a small amount of gas-phase breakthrough could be observed as the EESI-TOF is highly sensitive to gaseous compounds. 39 Nevertheless, overall our observations are consistent with the previous work of Tröstl et al. (2016) 29 which shows increasing contribution of higher volatility organics with increasing particle mass/size. Fig. 6 shows a more detailed comparison between the particle phase composition measured by the EESI-TOF, the predicted particle phase composition from the aerosol growth model and the gas phase measured by the PTR3 and the nitrate-CIMS. Since the condensation was modelled on a volatility-bin basis rather than compound-by-compound, the ratio of compounds corresponding to each bin in the condensed phase was assumed to be the same as in the gas phase. This was then multiplied by each integrated bin concentration to estimate the concentrations of each chemical species. Both the measured and modelled particle-phases are evidently less volatile and have higher average carbon oxidation state (OS C , calculated as 2 Â O/C -H/C) than the gas phase. When comparing the ultra-low volatility compounds (ULVOC) and ELVOC compounds where there were differences between the EESI-TOF data and the expected condensation of the gas-phase oxidation products, we observe that the modelled abundance of these compounds lies in a lower than predicted volatility. Lastly, fragmentation occurring in the PTR3 for such C 10 H x O 3-4 N 0-1 signals may be quite signicant, 42,43 resulting in lower observed gas-phase concentrations and therefore lower expected particle-phase mass concentrations from the aerosol growth model. Overall, we show that the particle composition measured by the EESI-TOF is in good agreement with the predicted particle composition from the condensation of gas-phase species, as well as that measured by the FIGAERO-CIMS. Uncertainties lie mainly in the compound-specic sensitivities in the EESI-TOF and the possibility of condensed-phase reactions taking place which have not been included in the aerosol growth model.

Conclusion

We presented a proof-of-concept application of the EESI-TOF at the CLOUD chamber at CERN, providing online measurements of the molecular composition of freshly nucleated nanoparticles from naphthalene and b-caryophyllene SOA. We offered a detailed intercomparison of the chemical composition measured with the EESI-TOF to the FIGAERO-CIMS. We demonstrated that the EESI-TOF can provide chemical composition measurement in real-time of particles as small as 20 nm and at mass loadings on the order of hundreds of ng m À3 , and therefore suitable for the study of growth under pristine atmospheric conditions. This has not been possible until now, as other online instruments such as the AMS or CHARON-PTR utilize aerodynamic lens-based systems with higher diameter cut offs. The FIGAERO and TD-DMA coupled to a CIMS are semi-continuous measurement techniques which may struggle to compete with the high time resolution required for nucleation experiments. By comparing our measurements to an aerosol growth model, we showed the consistency between the condensation of organic vapors and the measured particle phase. More work is required to fully understand observed differences, such as constraining the compound-specic response factors in the EESI-TOF or considering the possibility of condensed-phase processes in the aerosol growth model. The demonstrated capabilities of EESI-TOF enable the investigation of reactions occurring on rapid timescales, such as gas-to-particle partitioning or intra-particle reactions on a molecular level.

Paper

Environmental Science: Atmospheres