The SVN, illustrated in Fig. 1, creates an aerosol by ultrasonically nebulizing a small droplet placed on a thin film stretched across a water reservoir. The aerosol is then carried by a gentle flow of either house air (zero air, Aadco Instruments) or argon (Airgas, 99.999 % purity) into the AMS. The three main components of the SVN, described in detail below, are (1) a bottom cylinder with an ultrasonic transducer and water bath, (2) a thin film that is press-fit onto the top of the water bath by an upper cylinder with a slightly larger ID, and (3) a vertical glass tube that connects to the AMS. The connections between all components are airtight, but the apparatus is easily disassembled to inject samples onto the film, as well as to clean the thin film and change the water bath.

In the bottom section of the SVN, the 2.4 MHz ultrasonic transducer (Sonaer, Inc., model 241VM) is located just under the liquid reservoir, with a thin film stretched across the top of the reservoir to provide a clean nebulization surface for the sample. We use a 0.001 in. thick Kapton film or Teflon film, as these two were found to have the lowest background signal and the best performance in terms of the amount of aerosol generated compared to other materials tested. Press-fit onto the bottom piece is another PVC cylinder that has two side ports with carrier gas inlets and a larger hole in the top into which a 15 cm glass tube is seated. The distance from the thin film to the bottom of the glass tube is ∼1.5 cm. During experiments, the nebulized aerosol is carried up through the vertical glass tube and into the stainless steel tubing that leads to the AMS. Additional components such as Nafion^™ (Perma Pure LLC) dryers can be placed in line if desired, but such modifications were not investigated in the present work.

Samples can be introduced into the SVN using two different approaches: discrete injections of individual samples (for individual “one-shot” measurements) or continuous addition of a sample flow (for continual analysis, enabling signal averaging). For most studies, Milli-Q water was used as the solvent; in some cases we used HPLC-grade methanol, though the organic background signal is higher. This was likely due to a combination of increased organic background in organic solvents and incomplete evaporation of methanol prior to measurement. For most of the work described here, we used discrete injections of 2–5 µL of aqueous solutions manually deposited directly onto the center of the Kapton film. For continuous injections, solutions made with Milli-Q or organic solvents were introduced via a syringe pump (Harvard Apparatus model 22), which sends liquid flow (20–40 µL min-1) through a borosilicate capillary entering the SVN via a small downward-facing hole in the upper PVC piece (Fig. 1). In the future, such a port could also be used to provide automated discrete sample introduction using an autosampler.

For aqueous samples containing salts and small organic molecules, only 1 %–2 % of the original sample mass was observed to remain on the thin film after a discrete injection (Fig. S2). To ensure a clean surface between different samples, the surface was cleaned by nebulizing 2–8 µL of Milli-Q water off the surface 5–10 times over approximately 1 min. The cleanliness of the surface was then verified by nebulizing a salt solution (at least 0.5 g L-1) between each sample. The salt solution is necessary to ensure that any contaminants can be seen since pure water risks generating aerosol particles that are too small to be measured in the AMS. For samples in which carryover was observed (for example, the dissolved organic matter solutions discussed in Sect. 3.1), additional cleaning of the film was undertaken with sonication in a deionized water bath followed by rinsing with HPLC-grade methanol for >30 s. Careful maintenance of the surface ensures uncontaminated mass spectra and accurate quantification of the solution components.

While a number of different aerosol instruments could be used with the SVN, here we focus primarily on elemental analysis by the HR-ToF-AMS. The AMS has previously been described in detail (Canagaratna et al., 2007; Decarlo et al., 2006) and provides quantitative measurements of non-refractory material (organics, ammonium sulfate, ammonium nitrate, etc.) for aerosol particles between approximately 40 and 1000 nm. The mass spectrometer used in the AMS is a HR-ToF-AMS (Tofwerk AG), run under “V mode” for a mass resolution of 2000–3000 m/Δm. This mass-resolving power enables peak fitting and identification of all organic fragment ions observed here (m/z<130), which enables the calculation of quantitative elemental ratios for the organic mixture, after correcting for fragmentation bias during electron ionization (Aiken et al., 2007, 2008; Canagaratna et al., 2015). For AMS data collected using indoor or outdoor air, the intensities of CO+ and H2O+ are complicated by gas-phase interferences (N2+ and gas-phase H2O+). For samples compared to chamber or ambient online-AMS data sets, zero air was the carrier gas, standard empirical estimates were used, and the improved ambient method for elemental ratios was applied (Canagaratna et al., 2015). With the SVN, inert carrier gases such as argon can also be used, allowing for the direct measurement of the CO+ ion intensity. This is demonstrated below for dissolved organic matter; the majority of the other samples were run with zero air.

For discrete sampling, “fast MS” mode (Kimmel et al., 2010) was used because the pulse length of a single injection is ∼30–60 s long. Fast MS mode generates mass spectra every 0.5–2 s and the instrument cycles between the “closed” state, in which the aerosol beam is blocked, and the “open” state, in which the aerosol beam can reach the vaporization and ionization region for detection. For the work shown here, mass spectra were collected every 0.5 s for a duration of ∼15–18 s in the open state, followed by 3 s in the closed state. The closed spectrum provides information on the instrument background, including contributions from gas-phase species, and is subtracted from the open spectrum in data processing. For the high-resolution peak fitting and the analysis of the mass spectrum and the elemental ratios, the average mass spectrum across all injections is used. For quantification, the total signal under each injection pulse is used. For continuous injections, the standard AMS operating mode (“GenAlt” mode) was used. This provides an average mass spectrum (by subtracting the closed signal from the open signal), as well as particle time-of-flight (PToF) data (providing aerosol size distributions for all aerosol components), once per minute. All AMS data were analyzed using software packages Squirrel (v1.57I) and Pika (v1.16I), available at http://cires1.colorado.edu/jimenez-group/ToFAMSResources/ToFSoftware/ (last access: 25 July 2016).

The aerodynamic lens on the AMS has a transmission efficiency of nearly 100 % for particles with aerodynamic diameters of 70–500 nm; for somewhat smaller particles (40–70 nm), this transmission is lower but not negligible (Jimenez et al., 2003). Thus, high enough solution concentrations are used such that the dried particles formed in the nebulizer are larger than ∼100 nm in aerodynamic diameter. Collection efficiencies (CEs) in the AMS can vary depending on the extent to which aerosol particles bounce off the thermal element prior to vaporization (Docherty et al., 2013). This can impact the absolute concentrations observed, but for internally mixed samples, the relative concentrations of different aerosol components are independent of CE. In this work, most measurements (including elemental ratios) are reported as relative measurements, and thus no CE correction is applied. Some biases may arise if the aerosol is not internally mixed, but for all systems examined so far in PToF, no size dependence in composition was observed (Fig. S1 in the Supplement).

As described below, samples were prepared from a number of sources, including commercially available standards, the extracts of chamber and ambient aerosol particles collected on filters, and dissolved organic matter from the Pacific Ocean. For all solutions, either ultrapure water (18.2 M Ω cm, Milli-Q) or HPLC-grade methanol was used. Prior to use, all glassware was cleaned with a methanol solvent wash and baked at 450 ∘C for 6 h.

Chamber aerosol (enabling offline vs. online comparisons) was generated in the MIT 7.5 m3 Teflon environmental chamber, run in continuous-volume “semi-batch” mode. Details on the facility are given elsewhere (Hunter et al., 2014). Experiments were run at 20 ∘C, with <5 % relative humidity (RH), in the dark, and under low-NOx (<10 ppb) conditions using ozone as the oxidant. Ammonium sulfate seeds were added for an initial concentration of ∼ 60 µg m-3. The precursor, α-pinene, had an initial mixing ratio of 100 ppb; a Pen-Ray lamp (Jelight model 600) was used to add an initial ozone concentration of ∼ 700 ppb of ozone. The ozone concentration decreased due to consumption and dilution to 400 ppb by the end of the experiment. The initial organic loading was ∼ 70 µg m-3 and decayed due to dilution, sampling, and wall loss to a final value of ∼ 18 µg m-3. Filter samples were collected on Zeflour^® PTFE membrane filters (0.5 µm pore size) at flow rates of ∼5 L min-1 for 10 h. Laboratory blank filters were prepared by placing separate filters in the filter holder for 10 min before the start of the experiments. All filters were stored in baked aluminum foil packets, sealed in plastic bags, and placed in a freezer at -20 ∘C until extraction. Filters were extracted with ∼4 mL of HPLC-grade methanol. In order to avoid oxidation of the organic species in the extract, no sonication was used; instead, the vials were gently agitated by hand intermittently over 3 h. Solutions were concentrated by evaporating to dryness under a gentle stream of ultra-high-purity N2. Dried samples were stored in the freezer at -20 ∘C until reconstitution with Milli-Q water and analysis by the SVN–AMS. Blank subtraction was carried out with a scaling of the filter blank to 12 % of the sample signal, as determined from the internal standard in each sample.

Field samples from the Southern Oxidant and Aerosol Study (SOAS) in 2013 were collected on pre-baked Tissuquartz^™ filters (Pall Life Science, 8×10 in.) from Look Rock, TN, starting on 16 June 2013 using a high-volume aerosol filter sampler with a PM2.5 cyclone (Tisch Environmental, Inc.) as described by Budisulistiorini et al. (2015). For filter extraction, a 37 mm punch was extracted in a pre-cleaned scintillation vials with 20 mL high-purity methanol (LC-MS Chromasolv^® grade, Sigma-Aldrich) by sonication for 45 min. Filter extract was filtered through a 0.2 µm syringe filter (Acrodisc^® PTFE membrane, Pall Life Sciences) to remove suspended filter fibers. The filtered extract was then blown down to dryness under a gentle N2(g) stream at room temperature. An aerosol chemical speciation monitor (ACSM) (Ng et al., 2011a) was deployed at the same field site (Budisulistiorini et al., 2015); the average mass spectrum for the length of the filter sample was used for comparison with the present SVN–AMS measurements. For all analyses presented here (chamber and ambient) sufficient mass was extracted to enable the analysis of individual filter samples, with no combination of extracts from different samples required.

Standard solutions were prepared from commercially available compounds dissolved in Milli-Q water. Reagents used included ammonium sulfate, ammonium nitrate, isotopically labeled ammonium nitrate (NH415NO3), citric acid, mannitol, PEG-400, 4-hydroxy-3-methoxy-DL-mandelic acid (HMMA), and HPLC-grade methanol, all from Sigma-Aldrich, all at ≥99 % purity.

The dissolved organic matter (DOM) polysaccharide sample was collected at the Natural Energy Laboratory Hawaii Authority facility in Kona, Hawaii. Seawater from a depth of 20 m was pumped though a 0.2 µm filter to remove particles, and the high-molecular-weight fraction of organic matter in the filtrate was concentrated by ultrafiltration using a membrane with a 1 nm pore size and a nominal 1000 Da molecular weight cutoff. This fraction was desalted by serial dilution and concentration with Milli-Q water and then freeze-dried. Low-molecular-weight humic substances and residual salts were removed by stirring with anion (hydroxide form) and cation exchange resins (hydrogen form). The final product was freeze-dried to yield a fluffy white powder. Conventional CHNS analysis was carried out using a CE-440 elemental analyzer (Exeter Analytical). This powder was dissolved in Milli-Q water at approximately 1 g L-1 to prepare solutions for analysis.

Figure 2a shows a time series of measured aerosol mass concentrations of a typical nebulized aerosol pulse from a 4 µL solution containing approximately 0.33 g L-1 each of mannitol, ammonium sulfate, and ammonium nitrate. The nebulizer is turned on at t=0 and shortly afterwards (t=∼10 s) the aerosol packet is observed in the AMS. The start of the nebulization is timed so that a closed (background) measurement occurs during the downslope of the signal (closed measurements shown as gaps, e.g., t=∼16–21 s). This background is subtracted from the aerosol particle signal during data processing. Measurements are collected until the signal returns to the baseline (t=∼44 s).

Figure 2b shows the size distribution of the particles generated by nebulizing an aqueous solution of citric acid with continuous injection via syringe pump and a total concentration of ∼1 g L-1 into an SMPS (TSI, Inc.). The particles have size distributions centered at 150–200 nm (electrical mobility diameter). These particles were sampled into the SMPS without passing through a dryer. The SVN was approximately 3 m further away from the inlet of the SMPS so the particles are likely to be somewhat smaller than those entering the AMS, due to water evaporation in the dry carrier gas. We find injections of solutions with total concentrations above 0.2 g L-1 provide sufficient aerosol mass for analysis (Fig. S1). These measurements compare well with calculations based on the size of droplets reported by the manufacturer (Sonaer, Inc.) of approximately 1.7 µm using water solutions. Assuming that the density of the dried particle is 1.3 g cm-3 (Nakao et al., 2013), the minimum sample concentration that will form a 100 nm dried particle is approximately 0.3 g L-1. More dilute solutions do not generate signal in the AMS because the majority of the aerosol particles that are formed are too small for transmission through the aerodynamic lens of the AMS (Fig. 2b). To generate large enough aerosol particles from more dilute solutions, larger initial droplets could be formed by changing the transducer to one that vibrates at a lower frequency. However, for these larger droplets, drying will require the loss of a greater amount of solvent, so that any impurities in the solvent will make up a larger (and possibly even dominant) fraction of the resulting fine particles. Thus the use of ultrasonic nebulization at lower frequencies was not investigated here.

The primary goal of the SVN–AMS is to measure quantitative chemical information, specifically elemental ratios, from complex organic mixtures. We have characterized these for a number of different chemical systems, described below. Results are summarized in Fig. 5 (comparing SVN–AMS and online AMS mass spectra) and Table 1 (comparing elemental ratios measured with SVN–AMS with those measured by either online AMS or CHNS analysis).

One concern with using ultrasonic nebulization to generate aerosol particles is the possibility that the high temperatures possibly reached by the solution during nebulization may degrade the organic compounds, affecting their mass spectra (and hence measured elemental composition). Figure 5a shows a comparison of a solution containing 1 g L-1 of citric acid aerosolized with a TSI atomizer (TSI 3076, 276 kPa gas) (black) and the SVN (gold), with the inset showing a direct comparison among the intensities measured for each ion in the mass spectrum. The degree of agreement can be described by the dot product of the intensities for matching peaks in the two spectra, as well as the log of the intensities before taking the dot product (log–dot product), which gives the lower-intensity peaks greater weight. Very good overlap between the two mass spectra is observed, with a dot product of 0.99 and a log–dot product of 0.96. This indicates minimal degradation of the citric acid by ultrasonic nebulization.

A high degree of similarity is also observed between offline and online aerosol measurements for more complex mixtures. Figure 5b shows mass spectra for a comparison of offline (gold) vs. online (black) secondary organic aerosol (SOA), generated from the dark ozonolysis of α-pinene. The online mass spectra are the average real-time AMS mass spectrum averaged over the 10 h of filter collection. For all filter samples, spectra from the SVN are background subtracted using spectra collected from blank filter samples. These blanks provide the background for any trace organic material on the filters before collection as well as any background organic material introduced during sample preparation. The overlap in Fig. 5b between the mass spectra is very good, with a dot product of 0.98 and a log–dot product of 0.98. The elemental ratios are also very similar between the two samples with an H:C of 1.6 for both and O:C of 0.48 for the chamber and 0.49 for the SVN samples (Table 1). The largest difference is observed at m/z 44 (CO2+) and m/z 43 (C2H3O+) with a larger fraction of CO2+ in the offline sample. The intensity of CO+ (m/z 28) is also different, but only because it is set equal to the intensity of the CO2+ ion, as is commonly done for ambient sampling with the AMS (given that the CO+ ion generally cannot be distinguished from the much more abundant N2+ ion). The organic contribution from H2O+, OH+, and O+ is also constrained by the CO2+ signal so any differences in CO2+ intensity will also show up in those ions (Aiken et al., 2008). The observed difference in CO2+ and C2H3O+ ion intensity is likely a result of the extraction step prior to nebulization, which may preferentially dissolve the most water-soluble (oxidized) SOA components. Additionally, the online measurement is for fine mode aerosol while the offline measurement collects the full range of particle sizes on the filter. However, based on the agreement in H:C and O:C in the online and offline cases, these factors do not appear to bias elemental ratio measurements substantially.

Figure 5c shows a comparison of online and offline measurements of ambient organic aerosol, specifically ACSM measurements and SVN–AMS measurements of a filter extract collected simultaneously during the 2013 SOAS field campaign in Look Rock, TN (20:00 EST, 4 July to 07:00 EST, 5 July 2013). Since the ACSM is a unit-mass-resolution instrument, the HR-AMS data are degraded to unit mass resolution, and ions that are determined from the m/z 44 signal (m/z=15, 16, 17, 18, and 28) are excluded from the analysis. Additionally, ions at m/z 30 and 31 were removed from comparison because of interferences from the internal standard (m/z 31) and nitrate in the sample (m/z 30).

The two mass spectra in Fig. 5c have a high degree of agreement between the major ions (dot product of 0.98). However, there is substantially more variation among the two techniques than in the chamber study, especially in the lower-abundance peaks (m/z>45; see inset), as reflected in the lower log–dot product of only 0.90. Possible reasons for this lower correlation include fractionation from the extraction step, the different sizes measured (PM2.5 for the filter vs. fine mode aerosol for the ACSM) (Daellenbach et al., 2016), the uncertainty in ACSM signals at higher masses due to uncertainty in the relative ion transmission curve (Ng et al., 2011a), and the losses of relatively volatile compounds during collection, extraction, and handling. Additional work is necessary to quantify the importance of these effects, and care should be taken when comparing the full mass spectra for online compared to offline SVN–AMS analysis. The high degree of overlap in the intensities of the dominant ions between the online (AMS and ACSM) measurements and offline (SVN–AMS) results indicates that the ensemble organic composition for these aerosol samples is generally well-represented by the SVN–AMS measurements (Table 1). However, the estimated elemental ratios from a lower-resolution AMS are more uncertain than from the HR-ToF-AMS. Thus, the ratios for these samples in Table 1 are provided only as a demonstration of the overall agreement between the two techniques.

For the SVN, the small sample volume requirements can make it attractive for the analysis of other environmental samples that are soluble in water (or organic solvents) and that have low enough vapor pressures to remain in the condensed phase after sample preparation and nebulization. Here we demonstrate the analysis of the high-molecular-weight fraction of the polysaccharide fraction of dissolved organic matter (DOM) with the SVN–AMS. The DOM sample was prepared using a standard protocol for the isolation of this fraction of the organic material (see Sect. 2.3). This preparation removes the lower-molecular-weight compounds so chemicals such as methane sulfonic acid are not expected to be observed. Figure 5d shows an example AMS mass spectrum from DOM collected from the Pacific Ocean. The mass spectrum is dominated by oxidized fragments containing one or more oxygen atoms with smaller amounts of nitrogen-containing fragments. The sample preparation for the DOM removed all salts; thus the ammonium fragments were assumed to be from organonitrogen species and were assigned to the organic fraction. The measured N:C and H:C values of 0.081 and 1.7, respectively, match those measured by CHNS analysis (0.080 and 1.74, respectively). This demonstrates that with the SVN, microgram quantities of dissolved environmental mixtures can be nebulized and sampled into the AMS, providing a rapid, quantitative method to determine elemental ratios in these complex organic mixtures.