A description of the general experimental setup is given in more detail in Kirkby et al. (2011). The CLOUD chamber is a cylindrical stainless-steel container with an inner volume of 26.1 m3. It is filled principally with air (79 % nitrogen, 21 % oxygen) that is obtained from the evaporation of cryogenic liquids, with selected additional trace gases. Ozone (O3) concentrations in the chamber typically range from 200 to 1000 ppbv. RH typically varies between 21 % and 90 %, but is mostly held fixed between 37 % and 41 %. The trace gases sulfur dioxide (SO2) and NH3 are added on demand via individual independent lines. Fresh, humidified air and trace gases are fed into the chamber continuously at a total rate of 85 L min-1, while air is extracted by the measuring instruments. The desired concentration of each gas is achieved by continuous constant injection at the appropriate flow rate. The chamber is usually operated at an overpressure of 5 mbar to avoid contamination from outside the chamber. A pair of fans facilitate the mixing of the chamber contents (Voigtländer et al., 2012). The inside of the chamber is irradiated on demand by UV light from the top of the chamber (Kupc et al., 2011). This UV light induces photolytic reactions, in particular the oxidation of SO2 (at concentrations of 15 to 34 ppbv) to form H2SO4. The temperature inside the chamber is actively controlled and stable within 0.01 ∘C for the typical length of an experiment.

Some ionization always occurs inside the chamber via natural galactic cosmic rays. In addition, the chamber can be exposed to 3.5 GeV/c pions (π+) that are provided by the CERN Proton Synchrotron in one to three spills per minute. The intensity of the spills can be regulated, and the mean total ion pair production rate in the chamber is therefore adjustable between 2 cm-3 s-1 (π+ beam off) and 42 cm-3 s-1 (at the usual maximum available π+ beam intensity). An electrical clearing field of 20 kV m-1 can be applied by means of a pair of field cage electrodes, mounted at the top and the bottom of the chamber. This field will sweep out all ions in the chamber in about 1 s, providing an environment practically free of ions, when needed.

During the CLOUD 2 and CLOUD 3 campaigns, a wide array of instruments was arranged around the chamber, continuously analyzing its contents via 16 sampling probes. These sampling probes were mounted radially around the chamber and projected 0.5 m into the chamber. The instrumentation included an atmospheric pressure interface time-of-flight (APi-TOF) mass spectrometer to measure the chemical composition of ions (up to about 2 nm in size). Results from the APi-TOF are the main subject of this study, and the instrument is described below. The rest of the instrumentation included an Airborne Neutral cluster and Air Ion Spectrometer (NAIS) (Mirme et al., 2010), used to measure ions from 0.8 to 40 nm in mobility-equivalent diameter. A comprehensive suite of particle counting and sizing instruments facilitated aerosol number size distribution measurements, covering the range from 1.3 to 100 nm (Kirkby et al., 2011). A chemical ionization mass spectrometer (CIMS) (Kürten et al., 2011, 2012) measured H2SO4 concentrations down to about 105 cm-3 (4 × 10-3 pptv) at an accuracy of +100 %/-50 %. During CLOUD 3 only, a proton transfer reaction mass spectrometer (PTR-MS) (Norman et al., 2007) and a LOng Path Absorption Photometer (LOPAP) (Bianchi et al., 2012) were used to measure NH3 concentrations down to 35 pptv.

The APi-TOF sampled air from the CLOUD chamber using one of the radially mounted sampling probes. The sampling probe's inner diameter (ID) was 22.1 mm and its total length was 1.2 m, of which 0.5 m projected into the chamber. The APi-TOF shared the same sampling probe with the NAIS. The total sample flow from the chamber of 27.8 L min-1 (at -24 ∘C) to 34.5 L min-1 (at 19 ∘C) was split at 45∘ using a Y-splitter (Fig. 1). The flow from the Y-splitter to the APi-TOF (9.8 to 11.5 L min-1) was directed at the APi-TOF's orifice inlet, where 0.8 L min-1 was drawn into the instrument and the rest discarded.

Figure 1 shows that much of the sampling line was exposed to room temperature (> 5 ∘C). We thermally insulated the lines using Armaflex pipe insulation with aluminum tape on the outside, to minimize unwanted heating of the air taken from the chamber. The 0.8 L min-1 sample drawn into the APi-TOF was taken from the center of a ∼ 10 L min-1 flow, further mitigating heating of the sample. Simulations of the heat flux from warm ambient air into cool air flowing in a tube, insulated by a jacket of air, indicate that the APi-TOF sample may be heated up to several degrees before reaching the APi-TOF (e.g., from -25 to -16.5 ∘C or from 5 to 8.5 ∘C, ± 2 ∘C, at a room temperature of 20 ∘C). However, such heating would not qualitatively influence the conclusions regarding temperature effects in this study. In fact, it would only reduce the magnitude of the observed temperature effects.

The APi-TOF is a time-of-flight mass spectrometer built by Tofwerk AG and Aerodyne Research, Inc. A detailed description of the instrument and its capabilities is found in Junninen et al. (2010). The APi-TOF is designed to measure the mass-to-charge ratio of ambient ions of either positive or negative polarity. No ionization of the sample is performed, so only ions that are formed in the CLOUD chamber are detected. Air is sampled directly from atmospheric pressure via a critical orifice. In the interface (APi), ions are focused and guided through differentially pumped chambers to the time-of-flight mass spectrometer (TOF), where the pressure is reduced to 10-6 mbar.

During CLOUD 2 and CLOUD 3, the mass accuracy was better than 10 ppm. The resolving power (determined from the peak width at half maximum) was up to 4900 Th/Th (CLOUD 2) or up to 5300 Th/Th (CLOUD 3) for negative ions, and up to 4300 Th/Th for positive ions. The instrument was set to obtain mass-to-charge ratios up to either about 2115 Th (in positive mode and some experiments in negative mode) or 3300 Th (most experiments in negative mode). At all times during these measurements, the APi-TOF detected only singly charged ions; therefore, the unit thomson (Th) can also be thought of as atomic mass unit (u) or dalton (Da). To provide a comparison with condensation particle counters and mobility spectrometers, a singly charged ammonium bisulfate cluster ion at 3300 Th corresponds to about 2.1 nm in mobility-equivalent diameter, using the conversion procedure described by Ehn et al. (2011). The APi-TOF's ion transmission efficiency was set to have its maximum at about 900 to 1400 Th for negative ions, and at about 100 to 300 Th for positive ions. In the CLOUD campaigns, we recorded spectra at a time resolution of 5 s. The signal-to-noise ratio usually resulted in a maximum practical time resolution of 30 s.

Sampled ions may be subject to fragmentation inside the APi-TOF. Such fragmentation was mainly manifest by the usual near-absence of any H2O clustered with, for instance, sulfuric acid. The evaporation rate of H2O from these clusters is too rapid to survive detection in the non-equilibrium environment of the APi-TOF. However, many more strongly bound clusters can be detected, as will be shown here and has been shown before (e.g., Ehn et al., 2010; Junninen et al., 2010). Also, comparisons with ion mobility spectrometers demonstrated a good agreement with the APi-TOF's results (Ehn et al., 2011; Schobesberger et al., 2013a). Comparisons between the APi-TOF and the NAIS for our measurements produced a fair agreement as well, so the ion mass spectra obtained by the APi-TOF are, in general, representative of the actual population of small ions and ion clusters. However, a few molecules other than H2O may also be lost from some clusters during the sampling, as has also been suggested by comparisons between APi-TOF results and cluster simulations (Olenius et al., 2013b; Ortega et al., 2014).

The data obtained from the APi-TOF measurements were processed and analyzed using tofTools, a software package based on MATLAB and under continuous development, mainly at the University of Helsinki. Details on the analysis of APi-TOF data are found elsewhere (Schobesberger et al., 2013a).

The primary means of obtaining the gas-phase concentration of NH3 ([NH3]) were the results from the LOPAP (Bianchi et al., 2012). It was only available during CLOUD 3 and above 0 ∘C. Below 0 ∘C, measurements of [NH3] are available from the PTR-MS for some experiments in CLOUD 3. Ammonia concentrations could also be estimated from the calibrated mass flow controller settings.

In practice, [NH3] was directly measured whenever NH3 had been added during most of the CLOUD 3 campaign. Without the deliberate addition of NH3, values of [NH3] were below the detection limit of 35 pptv. More refined measurements during later campaigns showed that the contaminant level of [NH3] was in fact likely < 2 pptv for experiments at 5 ∘C (Almeida et al., 2013). The most plausible source of this contaminant NH3 was evaporation from the inside walls of the chamber. Therefore, we assumed that contaminant levels of [NH3] were 2 pptv at 5 ∘C and directly proportional to the desorption rate of NH3, assuming an activation energy of 33 kJ mol-1. Thus calculated contaminant levels of [NH3] ranged from 0.4 pptv (at -24.7 ∘C) to 4.1 pptv (at 19.8 ∘C).

Ammonia concentrations also had to be calculated for a selection of experiments below 0 ∘C when no direct measurement results of [NH3] were available. In the beginning of these experiments, [NH3] was above contaminant levels, but no NH3 was being added to the chamber. Therefore, a decay of [NH3] as measured previously by the LOPAP was used for our calculations, in addition to the proportionality to the desorption rate.

During the few experiments when NH3 was added during CLOUD 2, estimates for [NH3] were made using the settings of the mass flow controllers that control the gas flows into and out of the CLOUD chamber.

The same APi-TOF as in the CLOUD campaigns was deployed also at the Station for Measuring Ecosystem-Atmosphere Relations (SMEAR II) (Hari and Kulmala, 2005), where it measured negatively charged ions during spring 2011. The SMEAR II station is located in Hyytiälä, southern Finland, within a boreal forest. Tampere (population 213 000) is the closest larger town, 50–60 km southwest of the station. The station is the site of a host of continuing atmospheric observations, which includes extensive aerosol measurements that can be used to detect and analyze new-particle formation events (Kulmala et al., 2004a). For the results shown in this study, [NH3] was measured by a Monitoring instrument for Aerosols and Gases (MARGA) (Makkonen et al., 2014), and [H2SO4] was measured by a CIMS, similar to the one used at the CLOUD experiments.

The APi-TOF was situated inside a container in the forest, directly sampling ambient air in a setup similar to that used at the CLOUD chamber (details in Schobesberger et al., 2013a). It should be noted that the APi-TOF was tuned differently for those measurements, resulting in a reduced ion transmission efficiency at high m/z compared to the experiments at CLOUD.

During the CLOUD 2 and 3 campaigns, the conditions in the CLOUD chamber were set to precisely maintained conditions, specifically to a temperature of typically either 5 or 19–20 ∘C, an RH of typically 37 % to 41 %, an SO2 concentration between 15 and 34 ppbv, and an O3 concentration between 200 and 1000 ppbv. In the initial experiments, no NH3 was fed into the chamber.

A typical new-particle formation experiment in the CLOUD chamber was initiated by the UV lights being turned on, leading to a marked increase of [H2SO4] (Fig. 2a), which in turn triggered new-particle formation. The formation and subsequent growth of particles was measured by the particle or ion counting and sizing instrumentation, including the NAIS (Fig. 2b). For most of the investigated gas mixtures, the NAIS showed that ion-induced nucleation proceeded only or predominantly in negative polarity. Therefore the APi-TOF was mostly run in the negative mode (for detecting negatively charged ions) during both campaigns. Naturally, the main focus of this study also lies on negatively charged ions.

The APi-TOF measurements provide high-resolution mass spectra of ions and ion clusters up to about 2.1 nm in mobility-equivalent diameter, capturing exactly the critical first steps of the ion-induced pathway of new-particle formation (illustrated in Fig. 2b). The elemental compositions of ions are identified primarily by their exact mass. Therefore, it is advantageous to present mass spectra as mass defect diagrams (Fig. 2c). In such a diagram, the mass defect for each ion (i.e., the deviation from its nominal mass) is plotted against its mass-to-charge ratio. Any given ion will occupy a unique position in this diagram, and an addition of a specified atom or molecule will move an ion by a characteristic vector (e.g., see Fig. 2c insert).

Note that the APi-TOF spectra shown and analyzed in this study are averages over the duration of the steady-state conditions during a new-particle formation experiment (illustrated in Fig. 2). The steady-state periods were defined as the period during which no change in the APi-TOF ion spectrum occurred. Their duration ranged from 200 s to over 6 h.

The new-particle formation experiments in the CLOUD chamber covered a range in [H2SO4] levels from 3.3 × 106 to 1.4 × 109 molecules cm-3 (0.1 to 56 pptv). During a typical experiment, the dominant negatively charged ions were small sulfuric acid anion clusters, with the strongest signal, in most cases, from the trimer, (H2SO4)2 ⋅ HSO4- (Fig. 2c). Heavier ion clusters (> 350 Th; containing > 3 sulfur atoms) were considerably less abundant for most experimental conditions. These heavier clusters consisted mostly of H2SO4 molecules. However, they were observed not only as “pure” sulfuric acid clusters but also as clusters with base molecules, specifically molecules of NH3 or of various organic bases, mainly amines.

In general, larger clusters contained more base molecules. The predominant base in these clusters was NH3, yielding clusters of the form (NH3)m ⋅ (H2SO4)n ⋅ HSO4- (Fig. 2c). Only certain numbers of NH3 molecules (m) were seen for each number of H2SO4 molecules (n), depending on experimental conditions. This dependency will be discussed below in more detail. Note that neither NH3 nor amines had been deliberately fed into the chamber for these experiments. Instead, they were unintended impurities.

Some of the negatively charged clusters that grew by the addition of H2SO4 and base molecules had an additional oxygen atom (Fig. 2c). This can be explained by the growth starting from HSO5- instead of HSO4-. The role of HSO5-, as opposed to HSO4-, in the composition and growth of ion clusters, as well as its origin, will be described and discussed in Sects. 3.2 and 4.2.

Note that the APi-TOF did not detect any growing positively charged clusters under the typical experimental conditions discussed in this section; that is, when no NH3 was fed into the chamber, temperature was either 5 or 19 ∘C and RH was 40 %. This is consistent with simultaneous NAIS measurements, which did not show any growth starting from small positively charged ions (< 2.5 nm).

Practically all anion clusters that included the bisulfate ion (HSO4-) were also observed in the form where HSO4- was replaced either with HSO5- or, to a lesser extent, with SO5-. The HSO5- and SO5- ions appear to be less efficient than HSO4- in forming the initial clusters with H2SO4 molecules, based on the observed ratios of anion sulfuric-acid dimer signals to those of the bare anion, i.e.: [H2SO4⋅HSO4-]/[HSO4-]>[H2SO4⋅HSO5-]/[HSO5-]>[H2SO4⋅SO5-]/[SO5-]. The concentrations of HSO5- and SO5- in CLOUD were particularly high compared to the concentration of HSO4- when the concentration of H2SO4, the dominant precursor of HSO4-, was low. The relative amounts of charge carriers were also affected by the π+ beam intensity (i.e., the total ion concentration) and by the O3 concentration: higher beam intensity led to a higher fraction of HSO5- ions, whereas practically no HSO5- or SO5- was observed in experiments without O3 present in the CLOUD chamber. In addition, the abundance of HSO5--based ion clusters relative to HSO4--based clusters in CLOUD increased chiefly together with an increasing role of NH3. In the most extreme case – i.e., high [NH3], low [H2SO4] and high beam intensity – about 60 % of the large clusters (those containing 5–19 S atoms) were associated with HSO5-. A maximum of 7 % of the larger clusters were associated with SO5-, and a maximum of < 3 % with H2O11NS2-, probably in the form H2S2O8 ⋅ NO3-.

The cluster compositions were very similar regardless of which ion carried their charge. The most important difference between the different charge carriers was that we observed NH3 ⋅ (H2SO4)2 ⋅ HSO5- clusters, whereas the smallest ammonia-containing cluster associated with HSO4- was NH3 ⋅ (H2SO4)3 ⋅ HSO4-. The counts for NH3 ⋅ (H2SO4)2 ⋅ HSO5- clusters were usually more than an order of magnitude lower than the counts for similar clusters with one more H2SO4 molecule, NH3 ⋅ (H2SO4)3 ⋅ HSO5-, whereas the cluster NH3 ⋅ (H2SO4)2 ⋅ HSO4- was totally absent. Subsequently, the average ratio between the number of NH3 molecules (m) and the number of H2SO4 molecules (n) was initially higher in HSO5--based clusters than in the corresponding HSO4--based clusters. However, this difference decreased with an increasing cluster size, disappearing or staying approximately constant at about n≥9. The implications of these observations will be discussed in Sect. 4.2.

No amines were deliberately added into the chamber for the experiments discussed here, i.e., throughout the CLOUD 2 and CLOUD 3 campaigns. Amine contamination originated probably from the same source as NH3 (see Sect. 2.4). We can give some estimate of the contaminant levels of the dominant amine, C2H7N, based on measurements from later CLOUD campaigns when dimethylamine was also added into the chamber in several experiments (Praplan et al., 2012; Almeida et al., 2013). These estimates are based on direct measurements of dimethylamine concentrations down to 0.2 pptv performed during the later experiments and on measurements of the content of C2H7N in clusters seen by the APi-TOF. Based on those results, we speculate that gas-phase contaminant concentrations of C2H7N were between 0.1 and 1 pptv during the CLOUD 2 campaign, and about 0.1 pptv or even less during the CLOUD 3 campaign.

In the experiments discussed here, the highest abundance of the clusters containing contaminant organic bases (amines or amides) was usually seen on those clusters that contained a sulfuric acid tetramer anion. Tetramers were observed either without any base, such as (H2SO4)3 ⋅ HSO4-, clustered with NH3, or clustered with a basic organic compound. The organic base with the highest signal has the formula C2H7N (dimethylamine or ethylamine). Other bases observed in these clusters were CH5N (methylamine), CH4N2O (urea) and larger amines or amides (Fig. 3). Note that in some cases, we are unable to resolve whether one oxygen atom was part of the organic constituent or whether the ion was HSO5- instead of HSO4-. C2H7N was also seen to be bound to the sulfuric acid trimer anion, forming C2H7N ⋅ (H2SO4)2 ⋅ HSO4-, although with a signal about 2 orders of magnitude smaller than that of the cluster C2H7N ⋅ (H2SO4)3 ⋅ HSO4-. Notably, the corresponding cluster with NH3 instead of the amine, NH3 ⋅ (H2SO4)2 ⋅ HSO4-, was not observed, indicating its weaker base nature.

The clusters containing amines (or other organic bases) also evidently grew by the accretion of H2SO4 and NH3 molecules when amines were present at contaminant levels ([C2H7N] < 1 pptv): increasingly larger clusters of the type Y ⋅ (NH3)m ⋅ (H2SO4)n ⋅ HSO4- were formed, where Y was almost always one N-containing organic compound, and at maximum two such compounds ((C2H7N)2 or CH5N ⋅ C2H7N). In addition, the fraction of clusters that included N-containing organic compounds was smaller for larger clusters (n≥4) (Figs. 2c and Fig. 3).

In later experiments, NH3 was deliberately fed into the chamber to investigate new-particle formation over a range of [NH3] from contaminant levels (< 5 pptv; cf. Sect. 2.4) up to 1090 pptv. The range of investigated temperatures reached from +20 down to -25 ∘C. (Note the possibility of air sampled from the chamber at low temperatures being heated somewhat before reaching the APi-TOF, cf. Sect. 2.2.)

For all experimental conditions, negative ion clusters with more than 4 or 5 sulfur atoms grew by the accretion of NH3 and H2SO4 molecules, forming progressively larger (NH3)m ⋅ (H2SO4)n ⋅ HSO4- clusters. The number of NH3 molecules added on average per added H2SO4 molecule remained nearly constant from 4 or 5 sulfur atoms up to the upper detection limit of about 27 sulfur atoms, within the measurement uncertainties. These findings are illustrated in Fig. 4, which shows the average number of NH3 molecules (m) in clusters containing a certain amount of H2SO4 molecules (n), for each experiment and grouped by temperature.

We define the average number of added NH3 molecules per added H2SO4 molecule as Δm / Δn. This ratio corresponds to the slope of linear fits in m-vs.-n plots as in Fig. 4. For anions, we calculated values of Δm / Δn for n≥4, and found that Δm / Δn is well suited to describe the whole anion spectra during new-particle formation events in the NH3–H2SO4 system: two spectra with the same Δm / Δn were practically identical (unless Δm / Δn was close to zero, see Sect. 3.5), and, for a given temperature and RH, Δm / Δn was only dependent on the ratio between the NH3 and H2SO4 gas-phase concentrations, i.e., on [NH3] / [H2SO4] (color scale in Fig. 4, horizontal axis in Fig. 5). In our analysis for this study, values of Δm / Δn were calculated over the range 4≤n≤18 in the case of anion clusters, because Δm / Δn was approximately constant for n≥4 and we obtained a signal from clusters up to at least n=18 in most of the experiments.

At a given temperature and RH, the resulting Δm / Δn generally increased with an increasing value of [NH3] / [H2SO4], then flattened off when approaching the value of 1, and eventually reached a saturation value slightly larger than 1. At 19 ∘C, the maximum value of Δm/Δn of 1.1 to 1.2 was reached at the concentration ratio [NH3] / [H2SO4] ≈ 100 (Fig. 5). This concentration ratio was roughly coincident with the threshold for observing cation clusters (at 19 ∘C and 40 % RH) of the form (NH3)m ⋅ (H2SO4)n ⋅ NH4+ with m≈n and Δm/Δn≥1 (Figs. 4, 5). In an analogous way, the NAIS observed a formation of positively charged ions only when positively charged NH3–H2SO4 clusters were observable by the APi-TOF. Note that for the cation clusters, Δm/Δn was nearly constant from the monomer (n=1) onward and it was generally calculated over the range 1≤n≤17.

The relationship between Δm/Δn and [NH3] / [H2SO4] was similar under all experimental conditions, but the exact functional form of this relation was temperature dependent (Fig. 6a). For example, the value of Δm/Δn=0.2 was reached at [NH3] / [H2SO4] ≈ 0.1 when the temperature was 19 ∘C, but at [NH3] / [H2SO4] ≈ 0.7 when it was 5 ∘C, and at [NH3] / [H2SO4] > 1 when it was -25 ∘C. Also, the maximum observed values of Δm/Δn seemed to be reached at lower values of [NH3] / [H2SO4], and these maximum values were slightly higher at lower temperatures (e.g., a maximum Δm/Δn of 1.1 to 1.2 at 19 ∘C, versus a maximum Δm/Δn of 1.3 at -25 ∘C).

Note that in practically all our experiments with [NH3] / [H2SO4] < 1, only contaminant levels of [NH3] were present. These contaminant levels were not directly measured, but calculated under the assumptions described in Sect .2.4. In particular the temperature dependence of these low values of [NH3] is subject to those assumptions. Note also that most experiments at CLOUD were run with the RH between 37 % and 41 %, so the potential RH effects could not be thoroughly investigated. However, an increase of RH to > 68 % increased the value of Δm / Δn (Fig. 5). No significant effect on negatively charged clusters was observed due to RH changes in the range 30 % < RH < 60 %.

The role of NH3 in the formation of clusters became negligible at very low values of [NH3] / [H2SO4], i.e., when Δm/Δn dropped below 0.1 and the formation of pure binary clusters dominated (Fig. 6). Binary cluster formation is observed by the APi-TOF as pure sulfuric acid anion clusters, (H2SO4)n ⋅ HSO4-, because H2O molecules are lost in the sampling process (Sect. 2.3). The fraction of pure, NH3-free sulfuric acid clusters was calculated over the same cluster range as was used for calculating Δm/Δn (4≤n≤18), and like Δm/Δn, it was a function of [NH3] / [H2SO4]. The fraction of NH3-free sulfuric acid clusters increased with a decreasing value of [NH3] / [H2SO4], and such clusters were observed only below some temperature-dependent threshold value of [NH3] / [H2SO4] (Fig. 6b). Taken together, the presence of pure, NH3-free sulfuric acid clusters was favored by lower [NH3] / [H2SO4] ratios and lower temperatures.

Note that the NH3 contamination in the CLOUD chamber sets a lower limit to the minimum achievable level of [NH3]. Therefore, pure binary sulfuric-acid–water cluster formation can only be obtained at sufficiently high [H2SO4] level or at low temperatures. A low temperature both decreases the contaminant level of [NH3] (Sect. 2.4) and increases the threshold [NH3] / [H2SO4] below which binary cluster formation occurs (Fig. 6b).

Without adding any NH3 into the CLOUD chamber, most anions during a new-particle formation experiment in the CLOUD chamber were small sulfuric acid anion clusters. Heavier anion clusters, containing > 3 sulfur atoms, were not only “pure” sulfuric acid clusters but also clusters associated with base molecules, namely NH3 or various organic bases, mainly amines (Figs. 2c and 3). These findings agree qualitatively with observations made in the ambient atmosphere (Ehn et al., 2010), with independent measurements of such clusters in the laboratory (Bzdek et al., 2011), as well as with the results of simulations that consider classical collisions and evaporation rates based on quantum chemical methods (Olenius et al., 2013b). Both the earlier experimental and theoretical studies, as well as our results here, show that NH3 molecules can only cluster with anionic (= deprotonated) sulfuric acid clusters, i.e., form (NH3)m ⋅ (H2SO4)n ⋅ HSO4-, if n> 2. This lower size limit for the inclusion of NH3 molecules is explained by HSO4- itself acting as an electron donor (= Lewis base), in competition with regular bases such as NH3. Only when n> 2 is the cluster acidic enough to accept NH3 molecules. If NH3 or amines are available, their inclusion into larger clusters (n> 2) substantially enhances cluster stability, leading to a higher abundance as well as detectability in our measurements, compared to the corresponding pure sulfuric acid cluster (Ortega et al., 2014).

The observation of both ammonia and amines in these clusters here is remarkable because neither NH3 nor amines were deliberately fed into the CLOUD chamber for these experiments. All NH3 and amines were unintended impurities ([NH3] < 5 pptv, [C2H7N] < 1 pptv), yet they were found to play a crucial role in the chemistry of growing ion clusters.

In order of importance, the charge carriers in the observed base–sulfuric-acid clusters were HSO4-, HSO5-, SO5-, and H2O11NS2-. The HSO4- ion strongly correlated with [H2SO4] and was certainly formed by the de-protonation of H2SO4. A formation mechanism for SO5- has been investigated by Möhler et al. (1992). The initial step of this mechanism is the formation of SO3- from SO2 by the transfer of O- from O3-. Subsequently, SO5- forms from SO3- and O2 with a mediating N2 or H2O. We observed that O3 in the CLOUD chamber was required for the formation of both SO5- and HSO5-, supporting these or similar ion–molecule reactions as a source. Possible origins of the HSO5- ions were ion–molecule reactions similar to those probably leading to the formation of SO5-, a proton-transfer reaction with peroxymonosulfuric acid H2SO5, or the electronic charging of HSO5, while HSO5 itself had possibly been formed as described in Kurtén et al. (2009). H2O11NS2- was probably H2S2O8 ⋅ NO3-. H2S2O8 might have been formed from hydrated SO5 or from HSO5 (Kurtén et al., 2009). Nitric acid (HNO3) and its conjugate base NO3- were present as contaminants in the chamber.

By far the most abundant ions during the new-particle formation experiments at CLOUD that are discussed here were the clusters of type (NH3)m ⋅ (H2SO4)n ⋅ HSO4- and (NH3)m ⋅ (H2SO4)n ⋅ HSO5-. The latter type gains relevance in conditions of high [NH3]. Ammonia was present at n≥2 in the HSO5--based clusters but at n≥3 in the HSO4--based clusters. This suggests that sulfuric acid ion clusters will cluster more readily (i.e., build more stable clusters) with NH3 when the charge-carrying component is HSO5- instead of HSO4-. Most likely, this was the primary reason for the observed increase of the fraction of HSO5--based clusters at higher values of [NH3]. A secondary reason was the lower [H2SO4] during such experiments, which was usually coincidental. The only other difference to the HSO4--based clusters was an initially higher ratio m / n, as a direct consequence of the acceptance of NH3 molecules already at n=2. However, the composition of the clusters was otherwise the same, and the prevalent mechanism of cluster growth was not affected, namely the addition of a certain number of NH3 molecules per H2SO4 molecule. Therefore, we conclude that the presence and actions of charge carriers other than HSO4- do not alter the conclusions of this study. Interestingly, when n≥9, differences in the ratio m / n disappeared or became approximately constant, suggesting that at about that size the ion cluster had become large enough to render any chemical effect of its charge-carrying component unimportant.

In conclusion, we suggest that HSO5- is a somewhat weaker Lewis base than HSO4-. Therefore, clusters containing HSO5- more readily take up bases and less readily acids, as compared with clusters containing HSO4-, in line with all of our observations. Furthermore, the highest prevalence of HSO5- in all experiments at the CLOUD chamber to date has been in the dimethylamine–H2SO4 anion clusters that were produced during later experiments (in 2011 and 2012) following the addition of dimethylamine, a stronger base than NH3 (Almeida et al., 2013).

Throughout our experiments, contaminant levels of amines were below the limit of direct detection. We estimate the gas-phase contaminant concentrations of C2H7N between 0.1 and 1 pptv during the CLOUD 2 campaign, and about 0.1 pptv during the CLOUD 3 campaign, but they may have been even lower. The most abundant contaminant amine observed in anionic clusters was C2H7N, and the most abundant cluster containing C2H7N was C2H7N ⋅ (H2SO4)3 ⋅ HSO4- (Fig. 3). Note that C2H7N ⋅ (H2SO4)2 ⋅ HSO4- was observed, whereas NH3 ⋅ (H2SO4)2 ⋅ HSO4- was not. Nevertheless, the growth of the clusters at contaminant levels of amines proceeded almost exclusively by the addition of NH3 and H2SO4 molecules. The resultant dominant role of NH3 in the growth of the clusters, as opposed to organic bases (amines or amides), is most likely due to the differences in respective contaminant-level concentrations ([NH3] about 2 to 4 pptv, [C2H7N] < 1 pptv). Indeed, previous experimental studies on both positively and negatively charged dimethylamine–NH3–H2SO4 clusters showed that dimethylamine molecules would quickly displace NH3 molecules in these clusters already at low-pptv-level amine concentrations, whereas the opposite (displacement of dimethylamine by NH3 molecules) does not occur even at much higher gas-phase NH3 concentrations (Bzdek et al., 2010; Bzdek et al., 2011).

The effect of higher than contaminant gas-phase concentrations of amines, in particular of dimethylamine, on the composition of growing clusters and on particle formation rates was thoroughly investigated in subsequent CLOUD campaigns (Almeida et al., 2013; Bianchi et al., 2014). A large influence on cluster formation and particle formation rates was found at dimethylamine concentrations as low as 3 pptv. Specifically, growing ion clusters consisted of practically only dimethylamine and H2SO4, and particle formation rates were significantly enhanced. The enhancement of particle formation rates in those experiments was due to dimethylamine being a stronger base than NH3 and consequently forming more stable bonds with H2SO4 molecules, as has been shown both theoretically (e.g., Bzdek et al., 2010) and experimentally (e.g., Kurtén et al., 2008). The same reason can account for the observation (here, as in Almeida et al., 2013) that dimethylamine binds first to the sulfuric acid trimer anion, whereas NH3 requires one more H2SO4 in the cluster and binds first to the sulfuric acid tetramer anion. Specifically, dimethylamine competes more successfully than NH3 against HSO4-, which acts as a strong Lewis base in these clusters.

However, in the experiments presented here, no effect on the rate of new-particle formation was observed due to the contaminant levels of amines (< 1 pptv). Also, the composition of the growing ion clusters was dominated by NH3 and appeared to be unaffected by the occasional inclusion of an amine.

Interestingly, a similar dominance of NH3 over amines is also apparent in the composition of base–sulfuric-acid ion clusters that were observed during new-particle formation in the ambient atmosphere, specifically in the boreal forest environment in Hyytiälä, Finland (Kulmala et al., 2013; Schobesberger et al., 2013a). However, it was also shown that other oxidized organic compounds participate in those atmospheric new-particle formation events as well from an early stage on (Ehn et al., 2014), suggesting more complex mechanisms of formation and growth of clusters, which have not yet been determined in detail.

The APi-TOF measurements of NH3–H2SO4 clusters during particle formation experiments at the CLOUD chamber revealed that these clusters grow by the accretion of certain numbers of NH3 and H2SO4 molecules. The measurements covered the range up to clusters containing about 27 sulfur atoms. This maximum size corresponds to 2.1 nm in mobility equivalent diameter, when converted according to Ehn et al. (2011) using the bulk density of ammonium bisulfate (1780 kg m-3; and neglecting a likely involvement of H2O).

The anion clusters were mainly of the form (NH3)m ⋅ (H2SO4)n ⋅ HSO4-. The HSO4- ion acts as a strong Lewis base, so stable bonds with NH3 molecules are only possible for n≥3. For larger clusters (n≥4), the cluster growth was well characterized by the ratio of added NH3 molecules per added H2SO4 molecule, Δm / Δn, which we found to be dependent on the ratio [NH3] / [H2SO4] and temperature (Figs. 4, 6a). At sufficiently high values of [NH3] / [H2SO4], Δm / Δn saturated at just above unity. Note that a ratio of Δm/Δn=1 corresponds to the stabilization of each H2SO4 molecule by an NH3 molecule, as in ammonium bisulfate (whereas Δm/Δn=2 would correspond to the full neutralization of each H2SO4 molecule by two NH3 molecules, as in ammonium sulfate). All these observations make it very likely that the binding of molecules via strong hydrogen bonds between acidic and basic molecules and acid–base reactions were the dominant mechanism in both the initial formation of these clusters and their subsequent growth up to > 2 nm. The basic molecules were the HSO4- ion and NH3 molecules, while the acidic molecules were H2SO4 molecules. In terms of cluster composition, therefore, the chemical property of HSO4- appears to outweigh electrostatic effects due to the electric charge. Consequently, the enhancements of particle formation rates attributed to NH3 or to a negative charge (Kirkby et al., 2011) may both be the consequence of essentially the same process of acid–base stabilization (Kupiainen et al., 2012).

These findings are in agreement with previous studies that investigated the structure of the bonds in electrically charged or neutral NH3–H2SO4 clusters: both theoretical (e.g., Ortega et al., 2012; DePalma et al., 2012; Ortega et al., 2014) and experimental studies (e.g., Rozenberg et al., 2011; Froyd and Lovejoy, 2012; Johnson and Johnson, 2013) have shown that NH3 molecules are bound to H2SO4 molecules via the transfer of a proton from the acid to the base (acid–base reaction) in all but the smallest of these clusters. Note that for simplicity, the chemical formulas used in this paper disregard these reactions.

Note also that all of the observed clusters were probably hydrated before their H2O molecules were lost in the sampling process, due to the abundance of H2O at the conditions in the CLOUD chamber (e.g., Henschel et al., 2014). The stabilizing effect of H2O on H2SO4 is much smaller than that of NH3 (e.g., Kurtén et al., 2007a; DePalma et al., 2014), but at least in the absence of NH3, the contribution of H2O is important (e.g., Vehkamäki et al., 2002).

In Fig. 7 we summarize the characteristics of the observed composition of (NH3)m ⋅ (H2SO4)n ⋅ HSO4- anion clusters as a function of vapor concentration ratio [NH3] / [H2SO4] and temperatures. Pure binary H2SO4–H2O cluster formation was favored by low values of [NH3] / [H2SO4] and low temperatures. From a macroscopic point of view, this temperature dependence is consistent with the differences in the temperature dependences of the saturation vapor pressures of bulk H2SO4 and NH3. Namely, the saturation vapor pressure of H2SO4 decreases more steeply than that of NH3 with decreasing temperature (Vehkamäki et al., 2002; Hodgman, 1962); therefore the ratio of these pressures (psat.,NH3/psat.,H2SO4) increases with decreasing temperature.

From a molecular-level point of view, the influence of NH3 vs. H2SO4 vapors on the growing clusters is consistent with a barrier for addition of NH3 to the clusters but barrierless addition of H2SO4. This has also been deduced from surface-induced cluster dissociation and quantum chemistry (Bzdek et al., 2013). Specifically, the near-saturation of the NH3 content of the clusters at [NH3] / [H2SO4] > 10 suggests that roughly 10 collisions of NH3 with an under-neutralized cluster are necessary to add the base molecule to a growing cluster, or to overwhelm the evaporation rate of NH3 from the cluster. The presence of a barrier for the NH3 uptake would be expected to lead to a slower NH3 uptake at lower temperatures. Such a temperature dependence is consistent with the increased value of [NH3] / [H2SO4] required at lower temperatures to add NH3 molecules to under-saturated NH3–H2SO4 or H2SO4 clusters (Figs. 6, 7).

At high values of [NH3] / [H2SO4], the ratio Δm / Δn usually exceeded 1, approaching saturation (Figs. 6a, 7). This relationship appeared independent of the temperature within uncertainties and the investigated temperature range (-25 to 20 ∘C). However, our data set may not be sufficiently complete to resolve possible dependencies on temperature. In this regime of high values of [NH3] / [H2SO4], cation clusters of the form (NH3)m ⋅ (H2SO4)n ⋅ NH4+ were also observable, again with Δm/Δn≥1 (Fig. 6a). Note that the anion clusters tended to feature slightly higher saturation values of Δm / Δn at lower temperatures. This feature could be an indication of the enhanced evaporation of NH3 molecules from the clusters at higher temperatures, before or after the sampling. Indeed, results of computer simulations using the atmospheric cluster dynamics code (ACDC) suggest that there may be a systematic slight underestimation of the NH3 content of the experimentally observed clusters in these conditions of relatively abundant gas-phase NH3 that we cannot exclude (Olenius et al., 2013b; Olenius et al., 2013a). Those as well as earlier studies (e.g., Kurtén et al., 2007b) have demonstrated how at least small NH3–H2SO4 clusters are expected to hold onto a higher number of NH3 molecules at lower temperatures.

Since both the anion and cation clusters had Δm / Δn slightly above unity when [NH3] / [H2SO4] ≥10, the clusters became slightly less acidic as they grew. The mass defect diagrams of the corresponding ion mass spectra illustrate the details on the increase of the clusters' base content as they grew (Fig. 8a, b). The anion clusters were seen to grow first by the addition of acids due to the high basicity of HSO4-. From the tetramer onwards (#S ≥4; see Sect. 4.2 for exceptions), the anions became chemically (not electrically) neutralized, as the clusters grew by the addition of slightly more than one base per acid on average (Fig. 8b). For the cation clusters, NH3 was already present for a monomer (i.e., NH3 ⋅ H2SO4 NH4+), and the clusters became slightly more basic from the first step onwards (Fig. 8a). The maximum observed Δm / Δn ratios were about 1.4 for anion clusters and 1.1 for cation clusters. Therefore the NH3–H2SO4 molar ratio (m / n) exceeded unity beyond a certain cluster size, as illustrated in Fig. 9. Note that if the clusters' growth continued to adhere to the observed Δm / Δn ratios also beyond a mobility size of 2 nm, m / n would approach these values (grey curves in Fig. 9). And if Δm / Δn stayed below 1.5, as observed, also for larger clusters, ammonium bisulfate would remain the favored composition (as opposed to ammonium sulfate). However, it should also be noted here that for the positively charged clusters both Fig. 4 and, more clearly, Fig. 9 reveal an apparent discontinuity at m=n=10, with an increase of the slopes, specifically an increase of Δm / Δn from 1.0 to about 1.3 for the larger positively charged clusters.

Interestingly, we observed a wider range of the number of NH3 molecules m for any n in anion clusters than in cation clusters. We hypothesize that this observation follows from the requirement that NH3–H2SO4 clusters be basic enough to accept an additional proton (therefore becoming or staying positively charged), whereas they need to be acidic enough to donate a proton (therefore becoming or staying negatively charged). Sulfuric acid can form clusters with only itself (plus H2O) much more readily than can NH3. Therefore, highly acidic (NH3–)H2SO4 clusters can be formed, and they were indeed observed as anion clusters when [NH3] / [H2SO4] was sufficiently low (e.g., Fig. 2c). For these clusters Δm / Δn was < 1; that is, they continuously became more acidic as they grew. The possibility of such relatively acidic anion clusters leads to the wider range of observed m even at relatively high [NH3]. On the other hand, the maximum basicity of NH3–H2SO4 clusters (or their NH3 / H2SO4 molar ratio, m / n) is apparently limited, even at the highest investigated [NH3] (Fig. 6a). Only such relatively basic clusters can be observed as cation clusters.

Note that almost all of our experiments were performed with RH between 37 % and 41 %. Indeed, the data presented in Fig. 5 show that the observed NH3–H2SO4 clusters contained more NH3 at RH > 68 %. It remains a task for future studies to thoroughly investigate how the composition of these clusters changes with changing RH. It is also worth noting that the APi-TOF was set to measure positively charged clusters in only few experiments; only for experiments at 19 ∘C did these experiments include conditions with [NH3] / [H2SO4] < 100.

In this work we hypothesize that the growth mechanism of ionic NH3–H2SO4 clusters is governed chiefly by hydrogen bonds and ionic bonds formed by acids and bases and the availability of the gas-phase precursors NH3 and H2SO4. The electric charge somewhat increases the stability (Olenius et al., 2013a), but most importantly it provides an additional base in the form of HSO4-, whereas the NH4+ ion seems to exhibit neither strongly basic nor acidic behavior. At relatively high [NH3], both positively and negatively charged NH3–H2SO4 clusters grow by adding NH3 and H2SO4 molecules at a ratio Δm / Δn of about unity, once the cluster is large enough for the effect of the HSO4- or NH4+ ion to be neutralized in terms of its basicity or acidity (Figs. 4, 6a, 8a, 8b). Therefore we would expect electrically neutral clusters to grow by the same ratio Δm / Δn from the first bond formation onwards. At relatively low [NH3], only negatively charged clusters were observed in our experiments, growing mainly by the uptake of H2SO4 molecules (Figs. 4, 6a), whereas positively charged clusters were not observed. Electrically neutral NH3–H2SO4 clusters must have been formed under these conditions as well, since new-particle formation occurred without any ions in the CLOUD chamber (Kirkby et al., 2011).