We conducted experiments at the CERN CLOUD chamber (Cosmics Leaving Outdoor Droplets). With a volume of 26.1 m3, the chamber is built of electropolished stainless steel and equipped with a precisely controlled gas system. The temperature inside the chamber is measured with a string of six thermocouples (TC, type K), which were mounted horizontally between the chamber wall and the center of the chamber at distances of 100, 170, 270, 400, 650 and 950 mm from the chamber wall (Hoyle et al., 2016). The temperature is controlled accurately (with a precision of ±0.1 ∘C) at any tropospheric temperature between -65 and 30 ∘C (in addition, the temperature can be raised to 100 ∘C for cleaning). The chamber enables atmospheric simulations under highly stable experimental conditions with low particle wall loss and low contamination levels (more details of the CLOUD chamber can be found in Kirkby et al., 2011 and Duplissy et al., 2016). At the beginning of the campaign the CLOUD chamber was cleaned by rinsing the walls with ultra-pure water, followed by heating to 100 ∘C and flushing at a high rate with humidified synthetic air and elevated ozone (several ppmv) (Kirkby et al., 2016). This resulted in SO2 and H2SO4 concentrations that were below the detection limit (< 15 pptv and < 5 × 104 cm-3, respectively), and total organics (largely comprising high volatility C1–C3 compounds) that were below 150 pptv.

The air in the chamber is ionized by galactic cosmic rays (GCRs); higher ion generation rates can be induced by a pion beam (π+) from the CERN Proton Synchrotron enabling controlled simulation of galactic cosmic rays throughout the troposphere. Therefore, the total ion-pair production rate in the chamber is between 2 (no beam) and 100 cm-3 s-1 (maximum available beam intensity, Franchin et al., 2015).

The main instruments employed for this study were atmospheric pressure interface time-of-flight (APi-TOF, Aerodyne Research Inc. & Tofwerk AG) mass spectrometers. The instrument has two main parts. The first is the atmospheric pressure interface (APi), where ions are transferred from atmospheric pressure to low pressures via three differentially pumped vacuum stages. Ions are focused and guided by two quadrupoles and ion lenses. The second is the time-of-flight mass analyzer (TOF), where the pressure is approximately 10-6 mbar. The sample flow from the chamber was 10 L min-1, and the core-sampled flow into the APi was 0.8 L min-1, with the remaining flow being discarded.

There is no direct chemical ionization in front of the instrument. The APi-TOF measures the positive or negative ions and cluster ions as they are present in the ambient atmosphere. As described above, in the CLOUD chamber ions are formed by GCRs or deliberately by the π+ beam, leading to ion concentrations of a few hundred to thousands per cm3, respectively. In our chamber the dominant ionizing species are NH4+ and NO3- (see below). These ions mainly form clusters with the organic molecules, which is driven by the cluster energies. Therefore, the signals obtained do not provide a quantitative measure of the concentrations of the compounds. The higher the cluster energy with certain compounds, the higher the ion cluster concentration will be.

We calibrated the APi-TOF using trioctylammonium bis(trifluoromethylsulfonyl)imide (MTOA-B3FI, C27H54F6N2O4S2) to facilitate exact ion mass determination in both positive and negative ion modes. We employed two calibration methods, the first one by nebulizing MTOA-B3FI and separating cluster ions with a high-resolution ultra-fine differential mobility analyzer (UDMA) (see Steiner et al., 2014 for more information); the second one by using electrospray ionization of a MTOA-B3FI solution. The calibration with the electrospray ionization was performed three times, one for each temperature. These calibrations enabled mass/charge (m/z) measurements with high accuracy up to 1500 Th in the positive ion mode and 900 Th in the negative ion mode.

Additionally, two peaks in the positive ion mode were identified as contaminants and also used for calibration purposes at the three different temperatures: C10H14OH+ and C20H28O2H+. These peaks were present before the addition of ozone in the chamber (therefore being most likely not products of α-pinene ozonolysis) and were also detected by a proton transfer reaction time-of-flight mass spectrometer (PTR-TOF-MS). Both peaks appeared at the same m/z at all three temperatures. Therefore, based on the calibrations with the UDMA, the electrospray and the two organic calibration peaks, we expect an accurate mass calibration at the three temperatures.

All ambient ion composition data reported here were obtained during nucleation experiments from pure α-pinene ozonolysis. The experiments were conducted under dark conditions, at a relative humidity (RH) of 38 % with an O3 mixing ratio between 33 and 43 ppbv (Table 1). The APi-TOF measurements were made under both galactic cosmic ray (GCR) and π+ beam conditions, with ion-pair concentrations around 700 and 4000 cm-3, respectively.

Quantum chemical calculations were performed on the cluster ion formation from the oxidation products of α-pinene. The Gibbs free energies of formation of representative HOM clusters were calculated using the MO62X functional (Zhao and Truhlar, 2008), and the 6-31+G(d) basis set (Ditchfield, 1971) using the Gaussian09 program (Frisch et al., 2009). This method has been previously applied for clusters containing large organic molecules (Kirkby et al., 2016).

Under dry conditions (RH = 0 %) and GCR ionization, the main detected positive ions were N2H+ and O2+. With increasing RH up to ∼ 30 % we observed the water clusters H3O+, (H2O) ⋅ H3O+ and (H2O)2 ⋅ H3O+ as well as NH4+, C5H5NH+ (protonated pyridine), Na+ and K+ (Fig. 1a). The concentrations of the precursors of some of the latter ions are expected to be very low; for example, NH3 mixing ratios were previously found to be in the range of 0.3 pptv (at -25 ∘C), 2 pptv (at 5 ∘C) and 4.3 pptv (at 25 ∘C) (Kürten et al., 2016). However, in a freshly cleaned chamber we expect ammonia levels below 1 pptv even at the higher temperatures. For the negative ions, NO3- was the main detected background signal. Before adding any trace gas to the chamber the signal of HSO4- was at a level of 1 % of the NO3- signal (corresponding to < 5 × 10-4 molecules cm-3, Kirkby et al., 2016), excluding any contribution of sulfuric acid to nucleation in our experiments.

After initiating α-pinene ozonolysis, more than 460 different peaks from organic ions were identified in the positive spectrum. The majority of peaks were clustered with NH4+, while only 10.2 % of the identified peaks were composed of protonated organic molecules. In both cases the organic core was of the type C7-10H10-16O1-10 for the monomer region and C17-20H24-32O5-19 for the dimer region.

In the negative spectrum we identified more than 530 HOMs, of which ∼ 62 % corresponded to organic clusters with NO3- or, to a lesser degree, HNO3 ⋅ NO3-. The rest of the peaks were negatively charged organic molecules. In general, the organic core of the molecules was of the type C7-10H9-16O3-12 in the monomer region and C17-20H19-32O10-20 in the dimer region. For brevity we refer to the monomer, dimer (and n-mer) as C10, C20 (and C10n), respectively. Here, the subscript indicates the maximum number of carbon atoms in these molecules, even though the bands include species with slightly fewer carbon atoms.

Experiments at three different temperatures (25, 5 and -25 ∘C) were conducted at similar relative humidity and ozone mixing ratios (Table 1 and Fig. 3). Mass defect plots are shown for the same data in Fig. 4. The mass defect is the difference between the exact and the integer mass and is shown on the y axis versus the mass/charge on the x axis. Each point represents a distinct atomic composition of a molecule or cluster. Although the observations described in the following are valid for both polarities, the trends at the three temperatures are better seen in the positive mass spectra due to a higher sensitivity at high m/z.

The first point to note is the change in the distribution of the signal intensity seen in Fig. 3 (height of the peaks) and in Fig. 4 (size of the dots) with temperature. In the positive ion mode, the dimer band has the highest intensity at 25 and 5 ∘C (see also Fig. 1b), while at -25 ∘C the intensity of the monomer becomes comparable to that of the dimer. This indicates a reduced rate of dimer formation at -25 ∘C, or that the intensity of the ion signal depends on both the concentration of the neutral compound and on the stability of the ion cluster. Although the monomer concentration is higher than that of the dimers (Tröstl et al., 2016), the C20 ions are the more stable ion clusters as they can form more easily two hydrogen bonds with NH4+ (see Sect. 3.3). Thus, positive clusters formed from monomers may not be stable enough at higher temperatures. Moreover, charge transfer to dimers is also favored.

The data also show a “shift” in all band distributions towards higher masses with increasing temperature, denoting a higher concentration of the more highly oxygenated molecules and the appearance of progressively more oxygenated compounds at higher temperatures. The shift is even more pronounced in the higher mass bands, as clearly seen in the C40 band of the positive ion mode in Fig. 3a–c. In this case the combination of two HOM dimers to a C40 cluster essentially doubles the shift of the band towards higher mass/charge at higher temperatures compared to the C20 band. Moreover, the width of each band increases with temperature, as clearly seen in the positive ion mode in Fig. 4, especially for the C40 band. At high temperatures, the production of more highly oxygenated HOMs seems to increase the possible combinations of clusters, resulting in a wider band distribution.

This trend in the spectra indicates that the unimolecular autoxidation reaction accelerates at higher temperatures in competition to the bimolecular termination reactions with HO2 and RO2. This is expected. If unimolecular and bimolecular reactions are competitive, the unimolecular process will have a much higher barrier because the pre-exponential term for a unimolecular process is a vibrational frequency while the pre-exponential term for the bimolecular process is at most the bimolecular collision frequency, which is 4 orders of magnitude lower. Quantum chemical calculations determine activation energies between 22.56 and 29.46 kcal mol-1 for the autoxidation of different RO2 radicals from α-pinene (Rissanen et al., 2015). Thus, such a high barrier will strongly reduce the autoxidation rate at the low temperatures.

The change in the rate of autoxidation is also reflected in the O : C ratio, both in the positive ion mode (Fig. 4a–c), and the negative ion mode (Fig. 4d–f), showing a clear increase with increasing temperature. The average O : C ratios (weighted by the peak intensities) are presented in Table 2 for both polarities and the three temperatures, for all the identified peaks (total) and separately for the monomer and dimer bands. For a temperature change from 25 to -25 ∘C the O : C ratio decreases for monomers, dimers and total number of peaks. At high masses (e.g., for the C30 and C40 bands), the O : C ratio may be slightly biased since accurate identification of the molecules is less straightforward: as an example, C39H56O25 ⋅ NH4+ has an exact mass-to-charge ratio of 942.34 Th (O / C = 0.64), which is very similar to C40H60O24 ⋅ NH4+ at 942.38 Th (O / C = 0.60). However, such possible misidentification would not influence the calculated total O / C by more than 0.05, and the main conclusions presented here remain robust.

The O : C ratios are higher for the negative ions than for the positive ions at any of the three temperatures. Although some of the organic cores are the same in the positive and negative ion mode, the intensity of the peaks of the most oxygenated species is higher in the negative spectra. While the measured O : C ratio ranges between 0.4 and 1.2 in the negative ion mode, it is between 0.1 and 1.2 in the positive ion mode. An O : C ratio of 0.1, which was detected only in the positive ion mode, corresponds to monomers and dimers with two oxygen atoms. The presence of molecules with such low oxygen content was also confirmed with a proton transfer reaction time-of-flight mass spectrometer (PTR-TOF-MS), at least in the monomer region. Ions with O : C ratio less than 0.3 are probably from the main known oxidation products like pinonaldehyde, pinonic acid, etc., but also from minor products like pinene oxide and other not yet identified compounds. It is likely that these molecules, which were detected only in the positive mode, contribute only to the growth of the newly formed particles (if at all) rather than to nucleation, owing to their high volatility (Tröstl et al., 2016). In this sense, the positive spectrum could reveal both the molecules that participate in the new particle formation and those that contribute to growth. The differences in the O : C ratios between the two polarities are a result of the affinities of the organic molecules to form clusters either with NO3- or NH4+, which, in turn, depends on the molecular structure and the functional groups. Hyttinen et al. (2015) reported the binding energies of selected highly oxygenated products of cyclohexene detected by a nitrate CIMS, finding that the addition of OOH groups to the HOM strengthens the binding of the organic core with NO3-. Even when the number of H-bonds between NO3- and HOM remains the same, the addition of more oxygen atoms to the organic compound could strengthen the binding with the NO3- ion. Thus, the less oxygenated HOMs were not detected in those experiments, neither in ours, in the negative mode. The binding energies were calculated for the positive mode HOMs–NH4+ and are discussed in Sect. 3.3.

We also tested to which extent the formation of the C40 band could be reproduced by permutation of the potential C20 molecules weighted by the dimer signal intensity. Figure 5 shows the measured spectrum (blue) and two types of modeled tetramers: one combining all peaks from the C20 band (light gray) and one combining only those peaks with an organic core with O / C ≥ 0.4 – i.e., likely representing non-volatile C20 molecules (dark gray). The better agreement of the latter modeled mass spectrum of the tetramer band with the measured one suggests that only the molecules with O / C ≥ 0.4 are able to form the tetramer cluster. This would mean that C20 molecules with 2–7 oxygen atoms are likely not to contribute to the nucleation, but only to the growth of the newly formed particles. One has to note that the comparison of modeled and measured spectra relies on the assumption that the charge distribution of dimers is also reflected in the tetramers.

These two observations (change in signal distribution and band “shift”) are not only valid for positive and negative ions, but also for the neutral molecules as observed by two nitrate chemical-ionization atmospheric-pressure-interface time-of-flight mass spectrometers (CI-APi-TOF; Aerodyne Research Inc. and Tofwerk AG). This confirms that there is indeed a change in the HOM composition with different temperature rather than a charge redistribution effect, which would only be observed for the ions (APi-TOF). The detailed analysis of the neutral molecules detected by these CI-APi-TOFs will be the subject of another paper and is not discussed here.

A third distinctive trend in the positive mode spectra at the three temperatures is the increase in signal intensity of the C30 band at -25 ∘C. The increase in the signal of the trimer also seems to occur in the negative ion mode when comparing Fig. 3d and f. For this polarity, data from two campaigns were combined (Table 1). To avoid a bias by possible differences in the APi-TOF settings, we only compare the temperatures from the same campaign, CLOUD 10, therefore experiments at 25 and -25 ∘C. The increase in the trimer signal may be due to greater stability of the monomer–dimer clusters or even of three C10 molecules at low temperatures, as further discussed below.

Three points were addressed in the quantum chemical calculations to elucidate the most likely formation pathway for the first clusters, and its temperature dependence. These included (i) the stability of the organic cores with NO3- and NH4+ depending on the binding functional group, (ii) the difference between charged and neutral clusters in terms of clustering energies, and finally (iii) the possible nature of clusters in the dimer and trimer region.

The calculations showed that among the different functional groups the best interacting groups with NO3- are in order of importance carboxylic acids (R–C(= O)–OH), hydroxyls (R–OH), peroxy acids (R–C(= O)–O–OH), hydroperoxides (R–O–OH) and carbonyls (R–(R′–) C = O). On the other hand, NH4+ preferably forms a hydrogen bond with the carbonyl group independent of which functional group the carbonyl group is linked to; Fig. 6 shows examples of NH4+ clusters with corresponding free energies of formation for carbonyls (ΔG = -17.98 kcal mol-1), carboxylic acid (ΔG = -17.32 kcal mol-1) and peroxy acid (ΔG = -17.46 kcal mol-1). For the three examples shown, the interaction of one hydrogen from NH4+ with a C = O group is already very stable with a free energy of cluster ion formation close to -18 kcal mol-1.

To evaluate the effect of the presence of a second C = O on the binding of the organic compound with NH4+, we performed a series of calculations with a set of surrogates containing two C = O groups separated by different numbers of atoms, as shown in Fig. 7. The addition of a second functional group allows the formation of an additional hydrogen bond, increasing the stability of the cluster considerably (almost 2-fold) from about -18 to -34.07 kcal mol-1, whereby the position of the second functional group to form an optimal hydrogen bond (with a 180∘ angle for N–H–O) strongly influences the stability of the cluster, as can be seen in Fig. 7. Thus, optimal separation and conformational flexibility of functional groups is needed to enable effective formation of two hydrogen bonds with NH4+. This could be an explanation for the observation that the signal intensity is higher for dimers than for monomers, as dimers can more easily form two optimal hydrogen bonds with NH4+.

As shown by Kirkby et al. (2016), ions increase the nucleation rates by 1–2 orders of magnitudes compared to neutral nucleation. This is expected due to the strong electrostatic interaction between charged clusters. To understand how the stability difference relates to the increase in the nucleation rate, the ΔG of charged and neutral clusters were compared. For this, C10H14O7 and C20H30O14 were selected as representative molecules of the monomer and dimer region, respectively (Kirkby et al., 2016). Table 3 shows the calculated free energies of formation (ΔG) of neutral, positive and negative clusters from these C10 and C20 molecules at the three temperatures of the experiment. Results show that at 5 ∘C, for example, ΔG of the neutral dimer (C10 + C10) is -5.76 kcal mol-1 while it decreases to -20.9 kcal mol-1 when a neutral and a negative ion form a cluster (C10 + C10-). Similarly, trimers show a substantial increase in stability when they are charged – i.e., from -2.1 to -19.9 kcal mol-1, for the neutral and negative cases, respectively. The reduced values of ΔG for the charged clusters (positive and negative) indicate a substantial decrease in the evaporation rate compared to that for neutral clusters, and, therefore, higher stability. Comparing the NH4+ and NO3- clusters, the energies of formation for the monomer are -22.5 and -25.9 kcal mol-1, respectively, showing slightly higher stability for the negative cluster. Inversely, the covalently bound dimer showed greater stability for the positive ion (-30.9 kcal mol-1) compared to the negative ion (-25.6 kcal mol-1).

The temperature dependence of cluster formation is shown in Fig. 8 for the positive ion clusters. The blue and brown solid lines represent the needed ΔG for evaporation–collision equilibrium at 0.3 and 1 pptv HOM mixing ratio, respectively, calculated as described by Ortega et al. (2012). The markers show the calculated formation enthalpies ΔG for each of the possible clusters. For all cases, the trend shows an evident decrease in ΔG with decreasing temperature, with a correspondingly reduced evaporation rate.

At all three temperatures, the monomer cluster C10 ⋅ NH4+ falls well below the equilibrium lines, indicating high stability. Even though the difference between -25 and 25 ∘C is just -1.6 kcal mol-1 in free energy, it is enough to produce a substantial difference in the intensity of the band, increasing the signal at least 8-fold at -25 ∘C (as discussed in Sect. 3.2). In the case of the dimers, we consider the possibility of their formation by collision of a monomer C10 ⋅ NH4+ with another C10 (resulting in a C10C10 ⋅ NH4+ cluster) or the dimer as C20 ⋅ NH4+ cluster. The calculations show clearly that the cluster C10C10 ⋅ NH4+ is not stable at any of the three temperatures (green line). In contrast, the covalently bound C20 forms very stable positive and negative ion clusters (see Table 3). Trimers are mainly observed at lower temperatures. Since the C10C10 ⋅ NH4+ cluster is not very stable, we discard the possibility of a trimer formation of the type C10C10C10 ⋅ NH4+. Thus, the trimer is likely the combination of a monomer and a covalently bound dimer (C20C10 ⋅ NH4+). According to our calculations (Table 3) the preferred evaporation path for this cluster is the loss of C10 rather than the evaporation of C20. Therefore, we have chosen to represent only this path in Fig. 8. The ΔG of this cluster crosses the evaporation–condensation equilibrium around 5 and 14 ∘C for a HOM mixing ratio of 0.3 and 1 pptv, respectively, in good agreement with the observed signal increase of the trimer at -25 ∘C (Fig. 3a–c). It is important to note that, due to the uncertainty in the calculations, estimated to be ≤ 2 kcal mol-1, we do not consider the crossing as an exact reference.

The ΔG of the negative ion clusters, which are also presented in Table 3, decrease similarly to the positive ion clusters by around 2 kcal mol-1 between 25 and -25 ∘C. The cluster formation energies of the monomer and the dimer with NO3- are in agreement with the observed comparable signal intensity in the spectrum (Fig. 2) in a similar way as the positive ion clusters. The covalently bonded dimer ion C20 ⋅ NO3- is also more stable compared to the dimer cluster C10C10 ⋅ NO3-, suggesting that the observed composition results from covalently bonded dimers clustering with NO3- rather than two individual C10 clustering to form a dimer.

The formation of a covalently bonded trimer seems unlikely, so the formation of highly oxygenated molecules is restricted to the monomer and dimer region. The trimer could result from the clustering of C10 and C20 species. Similarly, and based on the C20 pattern observed in Fig. 1b, we believe that the formation of the tetramer corresponds to the collision of two dimers. No calculations were done for this case due to the complexity related to the sizes of the molecules, which prevents feasible high-level quantum chemical calculations.

Finally, a comparison of the ΔG values as presented in Table 3 confirms the expected higher stability of charged clusters compared to neutral clusters, decreasing the evaporation rate of the nucleating clusters and enhancing new particle formation.