The aerosol particles for the experiments were produced in the Manchester Aerosol Chamber (MAC). A detailed description of MAC can be found in Shao et al. (2021), and a brief introduction is given as below. The facility is run as a batch reactor, with an 18 m3 (3m(H)×3m(L)×2m(W)) fluorinated ethylene propylene (FEP) Teflon bag supported by three aluminium frames in which the upper and the lower frame can move freely to the expand or collapse as sampling air is introduced to or extracted from the chamber. The Teflon bag is enclosed inside a housing with temperature and relative humidity controlled by an air conditioning system. In total, two 6 kW Xenon arc lamps (XBO 6000 W/HSLA OFR; Osram) and a series of halogen bulbs, arranged in seven rows containing 16 bulbs each (Solux 50 W, 4700 K, and Solux MR16, USA), are mounted inside of the enclosure housing. The combination of five rows of halogen bulbs (two rows spare) and two Xenon arc lamps is chosen to be a good representative to mimic the solar spectrum in the wavelength of 290–800 nm (Alfarra et al., 2012). The calculated photolysis rate of NO2 (jNO2) from the O3–NO–NO2 photostationary state was (1.8–3) × 10-3 s-1 during the experimental period. The simulated irradiation spectrum intensity in our chamber is around a third of the measured solar spectrum at midday on a clear-sky day in Manchester during June (https://www.eurochamp.org/simulation-chambers/MAC-MICC, last access: 24 July 2021).

To ensure chamber cleanliness and data reproducibility, an automatic fill and flush cycle is conducted pre- and post-experiment using 3 m3s-1 purified clean air. In total, five to six cycles are routinely conducted between experiments, with the total number concentration of aerosol particles usually lower than 10 particles per cm-3 after cleaning. Air is scrubbed using a series of filters, including Purafil (Purafil Inc., USA) and charcoal to remove reactive gaseous compounds and a high-efficiency particulate absorbing (HEPA) filter (Donaldson Company, Inc., USA) to remove particles, and a dryer. To remove reactive compounds, the chamber is soaked in high concentrations of O3 (∼1 ppm) overnight between experiments, which is removed during pre-experiment fill and flush cycles on the subsequent day. An additional harsh cleaning procedure is conducted weekly, with high O3 (∼1 ppm) and UV for 4–5 h, to consume the remaining reactive compounds.

Seed particles, VOC, NOx, and water vapour are injected into chamber before illumination. Deliquescent ammonium sulfate (AS) seed (Puratronic; 99.999 % purity) are nebulised and introduced into a drum to mix before flushing into chamber. Liquid VOC precursors (α-pinene, isoprene, and o-cresol; Sigma-Aldrich; gas chromatography (GC) grade with ≥99.99 % purity) are injected, using a syringe, into a heated glass bulb in which the VOCs are instantaneously vaporised. The vaporised VOCs are flushed into the chamber with a flow of ∼0.5 bar high-purity nitrogen (electron capture device (ECD) grade; 99.997 %). Here, the experiments will investigate single and binary mixtures under modest NOx conditions (VOC/NOx of 3–10). NOx (mostly as NO2 in this study) is introduced through a cylinder with a flow of ∼0.3 bar high-purity nitrogen (ECD grade; 99.997 %), and the O3–NO–NO2 photostationary state will establish immediately after illumination. The detailed initial conditions for the chamber set up are shown in Table 1, and the experimental design will be described in detail in Sect. 2.3.

A series of instruments are equipped for gas-phase and particle-phase measurements in MAC. NO, NO2, and NOx are recorded by the NO–NO2–NOx analyser (model 42i; Thermo Fisher Scientific, USA) and O3 is measured by O3 analyser (model 49C; Thermo Electron Corporation, USA). RH and T inside of MAC are recorded by an EdgeTech dew point hygrometer (DM-C1-DS2-MH-13, USA) and two Sensirion SHT75 sensors (Farnell 413-0698, USA). The aerosol number size distribution (20–550 nm) is measured by a differential mobility particle sizer (DMPS). Hygroscopic growth factor (GF) at 90% RH of submicron aerosol particles (75–250 nm) was recorded by a custom-made hygroscopicity tandem differential mobility analyser (HTDMA; Good et al., 2010), which can be used to calculate the GF at given RH using a κ–Köhler approximation (Petters and Kreidenweis, 2007).

The particle bounce behaviour (bounce fraction; BF) was measured by a three-arm particle rebound apparatus with RH adjustment system. A brief instrumental description is provided below, and more details can be found in Bateman et al. (2014) and Liu et al. (2017), along with a schematic diagram in Fig. S1 in the Supplement of the latter. In total, three single-stage impactors operated in parallel in the system combined with a condensation particle counter (CPC; model 3772; TSI Incorporated, USA). The first impactor is not equipped with a plate (see step 1 in Fig. S1 in Liu et al., 2017), so particles can pass through the first impactor directly, which measures the total particle population (N1). The second impactor is equipped with a smooth plate (see step 2 in Fig. S1 in Liu et al., 2017), which provides a solid surface and allows particles to rebound from the impactor. The particle population measured after the second impactor represents the sum of particles that do not strike the impactor and that strike but rebound from the impactor (N2). The third impactor is equipped with a grease-coated plate (see step 3 in Fig. S1 in Liu et al., 2017). The coated grease is quite sticky, and all particles striking the plate will be stuck. Therefore, the particle number population after grease-coated plate provides a measure of the particles that do not strike the impactor (N3). The rebound fraction BF is defined in Eq. (1). During the experiments, a differential mobility analyser (DMA) was used to select monodisperse aerosol particles from chamber, with a mobility diameter of 100–200 nm, following the growth of the aerosol particles. The selected particle sizes are larger than the 50 % transmission diameter of the used impactor (84.9±5.4 nm; Bateman et al., 2014) to ensure a reliable bounce fraction measurement. 1BF=N2-N3N1-N3. BF can be used as a proxy of aerosol phase state, although it has a limited capability to represent semi-solid or solid particles over 100 Pas in viscosity (Bateman et al., 2015a; Reid et al., 2018). Nevertheless, it provides insights into the transition process between the liquid and solid or semi-solid phase, referred to as liquid-to-nonliquid transition below.

The chemical composition of the nonrefractory PM1 components (NH4+, NO3-, SO42-, and SOA) was recorded by a high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS; Aerodyne Research, Inc., USA). A detailed introduction of the instrument, calibration procedures, and quantification of the aerosol concentrations was described previously (DeCarlo et al., 2006; Jayne et al., 2000; Allan et al., 2003; Allan et al., 2004). The instrument was operated in “V mode” and recorded with a time resolution of 1 min (30 s mass spectrum (MS) + 30 s particle time of flight – PToF). The elemental ratio of O:C used for the proxy of organic oxidation state and the ions of NO+ and NO2+ used for the organic nitrate fraction calculation were derived from the high-resolution fitting on V-mode data.

Calibrations were performed before and after the campaign using monodisperse (350 nm) ammonium nitrate and ammonium sulfate particles, following the standard procedure in Jayne et al. (2000) and Jimenez et al. (2003). An averaged ionisation efficiency of ammonium nitrate was 9.38×10-8 ionsmolecule-1 from the two calibrations. According to the ion balance of ammonium nitrate and ammonium sulfate in the calibrations, the specific relative ionisation efficiencies (RIE) for NH4+ and SO42- are determined as 3.57±0.02 and 1.28±0.01, respectively. The RIE of all organic compounds used the default value of 1.4 (Alfarra et al., 2004). In this study, the AMS mass concentrations were corrected by comparing to the real-time DMPS unit mass multiplied by concurrent calculated density from AMS species. The mass ratio of AMS/(DMPS×density) is 0.4–1.0 in all investigated VOC systems.

The real atmosphere comprises a complex mixture of VOCs with various reactivities, many of which may act as SOA precursors with varying degrees of efficiency. Recent chamber studies have started using relatively simple VOC mixtures to investigate complex interactions in their ability to form SOA in the presence of inorganic seed particles (McFiggans et al., 2019; Shilling et al., 2019). Extending these previous studies, a project was designed that aimed at characterising chemical mechanisms, yield, and physicochemical properties (volatility, hygroscopicity, cloud condensation nuclei (CCN) activity, and phase behaviour) of SOA formed from initially iso-reactive biogenic and/or anthropogenic VOC photochemistry on ammonium sulfate seed and exploring the potential implications for the real atmosphere. Building on McFiggans et al. (2019), we added an anthropogenic VOC (o-cresol) into the initially designed binary mixtures of biogenic VOCs (isoprene and α-pinene). o-cresol can be emitted into atmosphere both directly, from biomass burning (Coggon et al., 2019; Koss et al., 2018), and indirectly, from the oxidation of toluene (e.g. motor vehicles and solvent use; Fishbein, 1985). Therefore, anthropogenic sources are one of the main contributors to o-cresol, but it is worth noting that natural biomass burning can also be an important contributor. o-cresol is chosen by virtue of its comparable ⋅OH reactivity with the chosen biogenic VOCs (Coeur-Tourneur et al., 2006), which enables the comparable initial concentration to have an equal reactivity with ⋅OH at the beginning of the experiment (referred as iso-reactive). The modest SOA yield of o-cresol (Henry et al., 2008) gives a good contrast with the low-yield isoprene and high-yield α-pinene. The overall experiment design enables us to explore the initial SOA formation in iso-reactive single, binary, and ternary VOC mixtures oxidation (details can be found in Voliotis et al., 2021). This paper focuses on the aerosol phase state of seeded SOA from iso-reactive single and binary VOCs photochemistry. Unfortunately, the bounce impactor was unavailable for the ternary mixture experiment.

Figure 1 shows the rebound curves of the 100–200 nm multicomponent aerosol particles formed in various VOC systems for the period when the organic mass fraction in NR–PM1 is larger than 0.05. For all investigated VOC systems, the aerosol particles exhibited BF > 0.8 at RH < 30 % and BF < 0.2 at RH > 80 % at room temperature (18 ∘C). Between 30 % and 80 % RH, the BF monotonically decreased with the increasing RH, indicating a gradual transition in BF (usually within 15 %–25 % RH width for BF declining from >0.8 to <0.2). Assuming the aerosol particles to be nonliquid if their BF >0.8 and liquid if BF < 0.2 (Bateman et al., 2015a; Liu et al., 2017) implies a gradual transition with RH in all investigated VOC systems, which is in contrast to the rapid dissolution corresponding to the deliquescence of inorganic salt particles (Tang and Munkelwitz, 1993; Kreidenweis and Asa-Awuku, 2014). Owing to limitations of the technique in differentiating particle viscosity at high values (Bateman et al., 2014), the nonliquid phase could represent the semi-solid or solid phase. In addition, the rebound curves varied along with the SOA formation and subsequent evolution as the photochemistry continues in all the investigated VOC systems (as shown in Fig. 1). To illustrate the influences of chemical composition, the overview of the rebound curves as a function of organic–inorganic mass ratio (MRorg/inorg) in all VOC systems is shown in Fig. S1, indicating the potential important role of MRorg/inorg (and maybe the SOA composition) in determining the phase behaviour as RH.

To clearly describe the phase behaviour during SOA formation evolution, the RHBF=0.2, RHBF=0.5, and RHBF=0.8 are determined, representing the RH at which the BF is close to 0.2 (±0.15), 0.5 (±0.15), and 0.8 (±0.15). It is worth noting that the determination of RHBF=0.2,0.5,0.8 carries a maximum error of ±5 %, owing to the measurement resolution and resultant number of data points. By tracking the variation in RHBF=0.2,0.5,0.8 during SOA formation evolution, we can understand the tendency of the phase behaviour change and gain insight into the liquid-to-nonliquid transition for the variation in RH at which BF changes from 0.8 to 0.2 among VOC systems. As different VOCs have different reactivity with the oxidant and yield, the formed SOA mass after 6 h photo-oxidation and, consequently, the MRorg/inorg were different among VOC systems. For o-cresol/isoprene, 50 % reactivity o-cresol, o-cresol, α-pinene/isoprene, α-pinene/o-cresol, 50 % reactivity α-pinene, and α-pinene systems, and the MRorg/inorg reached up to 0.53±0.02, 1.27±0.05, 2.47±0.06, 3.77±0.11, 3.77±0.01, 4.45±0.04, and 7.37±0.06 for the last rebound curve at the end of experiments, respectively. As shown in Fig. 2, the RHBF=0.2,0.5,0.8 increased towards a maximum as an increase in the MRorg/inorg in all investigated VOC systems, indicating an increase in rebound tendency during the SOA evolution on deliquescent sulfate seed. Figure 2a–c showed a co-increasing trend between RHBF=0.2,0.5,0.8 and MRorg/inorg as MRorg/inorg<2; thereafter, RHBF=0.2,0.5,0.8 remained constant at 70 %–75 %, 65 %–70 %, and 50 %–65 %, respectively, in all VOC systems. Herein, the reason for the decrease in RHBF=0.5 from 60 % to 50 % at MRorg/inorg∼1 in the o-cresol system was not known, and the same behaviour was not observed for RHBF=0.2,0.8.

As expected, the change in MRorg/inorg during SOA formation and evolution differed in various VOC systems (as shown in Fig. 3), depending on their SOA production rate (as shown in Fig. S2) and noting the comparable initial sulfate seed concentrations. The order of the SOA production rate (from low to high) in all the investigated VOC systems was o-cresol/isoprene < 50 % reactivity o-cresol < o-cresol < α-pinene/isoprene < α-pinene/o-cresol < 50 % reactivity α-pinene < α-pinene (note that insufficient mass was generated from the isoprene system). It is worth noting that the SOA production rate in α-pinene/isoprene, α-pinene/o-cresol, and 50 % reactivity α-pinene systems are very similar, and their order is determined by the minor difference in the slopes in Fig. S2. Clearly, a higher SOA production rate increased the MRorg/inorg faster, causing the phase behaviour change. As shown in Fig. 4, RHBF=0.2,0.5,0.8 showed an increasing trend towards a maximum across the various VOC systems, which is highly related to how fast the SOA were formed. With an increasing SOA formation rate from the lowest o-cresol/isoprene to the highest α-pinene system, a shorter time was taken for RHBF=0.2,0.5,0.8 to reach the maximum. Take RHBF=0.5 as an example; it can be seen from Fig. 1 that the RHBF=0.5 at the beginning of the photochemistry was lower than 40 % for all investigated VOC systems. Figure 4b shows that, for the o-cresol/isoprene system with the lowest SOA formation rate, it took ∼3.5 h for RHBF=0.5 to increase to 50 % and ∼5 h to approach 60 %. For comparison, for α-pinene/isoprene (the moderate) and α-pinene (the highest) systems, it takes only ∼1.5 and 0.5 h for the RHBF=0.5 to increase to 50 % and 70 %, respectively. Similar results can also be found for the cases in RHBF=0.2,0.8.

It can be seen that, in all of the investigated VOC systems, when the SOA mass fraction was >0.05, the phase dependence on MRorg/inorg was qualitatively similar, irrespective of the single (binary) biogenic (anthropogenic) SOA precursors. That is, the RHBF=0.2,0.5,0.8 increased towards a maximum value with an increase in the MRorg/inorg during the SOA formation evolution. With regard to the SOA from biogenic or anthropogenic VOCs, oxidation showed an amorphous solid property rather than the expected liquid phase as deliquescent inorganic particles (Bateman et al., 2015a; Virtanen et al., 2010; Saukko et al., 2015; Saukko et al., 2012), such as the RHBF=0.2,0.5,0.8 of SOA formed from α-pinene and toluene photo-oxidation, were 80 %–90 %, 80 %–85 %, and ∼65 %, respectively (Bateman et al., 2015a). It is expected that the increasing aerosol rebounding tendency can happen with increasing SOA mass condensing on the deliquescent sulfate seed. This speculation is supported by our results which show an increase in RHBF=0.2,0.5,0.8 toward the maximum with more SOA condensation. Additionally, the maximum RHBF=0.2,0.5,0.8 at high MRorg/inorg of ∼8 in our study was 10 %–15 % lower than SOA mentioned above, indicating that the presence of a small mass fraction of inorganic compounds (∼10 %) makes multicomponent aerosol particles bounce less than the SOA. Moreover, the time taken for RHBF=0.2,0.5,0.8 to reach the maximum depended on how rapidly the SOA was formed, i.e. there was a shorter time for the faster SOA production rate of the investigated VOC systems. This general behaviour was independent of the yield of the VOC and whether the precursor was biogenic or anthropogenic. In addition, the individual VOC systems behaved in the same way as the VOC mixtures.

As shown in Fig. 3, it should be noted that, to avoid a high signal-to-noise ratio (SNR) for low organic mass loading measured by HR-ToF-AMS, the data points with an organic mass fraction larger than 0.05 (assuming uniform chemical composition) were selected for consideration. It was observed that particulate nitrate (maximum 5 %–16 % of NR–PM1) was formed during photochemistry in all investigated VOC systems. Nitrate can either be formed from the oxidation of NO2, followed by the neutralisation of NH3, or an organic oxidation product (organic nitrate). The organic nitrate fraction in the total nitrate signal can be estimated following the NO+/NO2+ ratio method proposed by Farmer et al. (2010), given the differentiation of the NO+/NO2+ ratio for pure NH4NO3 (2.6; from calibrations) and organic nitrate of 10–15 (Bruns et al., 2010; Fry et al., 2009; Kiendler-Scharr et al., 2016; Reyes-Villegas et al., 2018). As shown in Fig. S3, the organic nitrate fraction in the total nitrate signal was lower than 20 % in almost all investigated VOC systems, except for the last 2 h of o-cresol system (up to ∼26 %). This NO+/NO2+ ratio method can indicate organic nitrate statistically only if the organic nitrate fraction is larger than 15 % (Bruns et al., 2010). Thus, the observed particulate nitrate is mainly inorganic nitrate with a small contribution of organic nitrate (<20 %) for all VOC systems in this study. The small contribution of organic nitrate mass has little influence on the onset of the organic mass fraction >0.05. Moreover, Li et al. (2017) found that the pure NH4NO3 particles adhered on the impactor even though the RH had been reduced to ∼5 %. It is worth noting that the variable particulate nitrate across all VOC systems in this study might have some influence on the phase behaviour of the multicomponent aerosol particles during SOA formation evolution, and the influence of changing inorganic component ratios on the BF in multicomponent mixtures should be the focus of this work.

In addition to the control of the BF by MRorg/inorg, there is an indication that the SOA composition across the various VOC systems may also impact the aerosol phase state. As indicated in Fig. 2b and c, the increase in the RHBF=0.5,0.8 as a function of MRorg/inorg is less rapid in the α-pinene/isoprene, 50 % reactivity α-pinene, and the α-pinene system (referred to as biogenic volatile organic compound (BVOC) systems, which are anthropogenic free) than the anthropogenic VOC (AVOC) systems (AVOC-containing systems, including o-cresol, 50 % reactivity o-cresol, o-cresol/isoprene, and α-pinene/o-cresol). It can be seen that, for the AVOC-containing systems, the RHBF=0.5 and RHBF=0.8 were 65 %–70 % and 40 %–55 % when MRorg/inorg approached ∼1, whereas there was only 40 %–45 % and ∼25 % in the 50 % reactivity α-pinene and α-pinene/isoprene systems (BVOC systems). Interestingly, with more SOA condensing, the discrepancy disappeared, and the RHBF=0.5 and RHBF=0.8 converged for BVOC and AVOC-containing systems as the MRorg/inorg>2. In contrast, the RHBF=0.2 as a function of MRorg/inorg was the same for BVOC and AVOC-containing mixtures. This indicates that the decrease in RH to achieve the liquid-to-nonliquid transition for BVOC and AVOC-containing systems was different at moderate MRorg/inorg of ∼1 but quantitatively similar at low and high MRorg/inorg.

The above evidence indicates that the VOC type (BVOC or AVOC-containing systems), and then SOA composition, can play a second-order role in the phase behaviour of multicomponent aerosol particles. Here, we used degree of oxidation of the SOA (O:C atomic ratio derived from HR-ToF-AMS) as a proxy of SOA composition to explore its relationship with phase behaviour. As shown in Table 1, for BVOC systems the averaged atomic O:C ratio of SOA in α-pinene/isoprene, 50 % reactivity α-pinene, and α-pinene systems was 0.43±0.05, 0.45±0.04, and 0.36±0.03, respectively. For AVOC-containing systems, the O:C ratio of SOA was 10 %–50 % higher than BVOC systems, with 0.69±0.05, 0.66±0.05, 0.64±0.03, 0.48±0.05 in o-cresol/isoprene, 50 % reactivity o-cresol, o-cresol and α-pinene/o-cresol systems, respectively. In addition, the RHBF=0.2,0.5,0.8 response to the MRorg/inorg was coloured according to the O:C ratio in Fig. 2. The O:C ratio was characteristic for the SOA precursors and showed little variation through individual experiments. No direct relationship between O:C ratio and the phase behaviour change was observed during the SOA formation evolution among all investigated VOC systems. Previous studies have shown a tentative dependence on the O:C ratio of the organic particles with phase behaviour (e.g. glass transition temperature; Koop et al., 2011; Shiraiwa et al., 2017). In contrast, Saukko et al. (2012) showed that the O:C ratio only influenced rebounding behaviour of SOA formed from n-heptadecane in a potential aerosol mass (PAM) reactor, with no influence in the α-pinene, longifolene, and naphthalene systems. There was some evidence showing that aerosol morphology (single well-mixed phase or phase separation) was closely related to the O:C ratio of SOA and the organic–inorganic mass ratio (Bertram et al., 2011; Krieger et al., 2012; You et al., 2014; Freedman, 2017; Song et al., 2012; Smith et al., 2013). It is unknown whether the morphology plays a role in the phase behaviour discrepancy between BVOC and AVOC-containing systems at moderate MRorg/inorg, considering their O:C difference; nevertheless, this is of interest and needs further investigation.

As we know that the chemical composition and RH are key factors influencing aerosol water uptake at a given size (Kreidenweis and Asa-Awuku, 2014). To discuss the mixing role of the chemical composition and RH on BF, the GF at the RH of the BF measurement was calculated from the HTDMA measurement using the κ–Köhler equation, and the relation between the BF and the GF was plotted in Fig. 5. It can be seen that the BF showed a monotonic decrease from ∼1 to ∼0 with the increasing GF from ∼1 to ∼1.15 in all VOC systems. As the GF is larger than 1.15, the aerosol particles are kept in the liquid phase (BF ∼ 0). This evidence indicated the key role of increasing the aerosol water uptake in the phase transition from nonliquid to liquid. The BF–GF relation of the varying multicomponent aerosol particles, including SOA and inorganic compounds, is comparable with a previous study (Bateman et al., 2015a). They measured the BF of the SOA from α-pinene, toluene, and isoprene and found that the BF is nearly 0 when the GF is larger than >1.15 (Bateman et al., 2015a).