The isothermal evaporation behavior of the SOA particles is illustrated by the VFR as a function of tR in Fig. 1. The evaporation rate of the dry SOA particles was the slowest, and the differences in the sum (Fig. 2) and factor thermograms (Figs. 3 and 4) between two evaporation stages were minor. The particle evaporation rate became faster with increasing RH for both SOA systems. When particulate water was present, the contribution of compounds in the SVOC range was reduced during fresh stages (Fig. 2). As shown in previous studies (Yli-Juuti et al., 2017; Buchholz et al., 2019; Li et al., 2019; Zaveri et al., 2020), considerable kinetic limitations exist for the evaporation of volatile compounds in this type of dry SOA particles due to the substantially high viscosity. Particulate water reduces the viscosity and thus enhances particle evaporation with increasing RH. The comparable evaporation rates under intermediate- and high-RH conditions suggest that particle evaporation can be approximated as a liquid-like process for both conditions (i.e., at RH ≥ 40 %), but in addition to this plasticizing effect, particulate water content may also induce aqueous-phase processes during isothermal evaporation (Buchholz et al., 2019; Petters et al., 2020). For the investigated SOA particles, we observed strong evidence of such processes under high-RH conditions (RH = 80 %). These are detailed in Sect. 3.3.3. Quantifying the effects of particle viscosity and aqueous-phase processes on the SOA particle evaporation would require developing detailed process models considering particle phase chemistry, which is not the primary focus of this study.

At any set RH, the evaporation rate of the SQTmix SOA particles was slower than that of the α-pinene ones, although both SOAs were produced under comparable oxidation conditions. Such distinguishable evaporation patterns are most likely driven by (i) the distinct particulate volatility distributions jointly controlled by molecular weight and functionality, expressed by elemental composition as a proxy (Li et al., 2016), and/or (ii) the possible differences in particle bulk viscosities especially under dry conditions.

In Fig. 2, the thermal desorption behaviors of particle samples which were collected at fresh (avg. tR=0.25 h) and RTC (avg. tR=4.25 h) evaporation stages under dry (RH < 7 %) and high-RH (RH 80 %) conditions are displayed as normalized STG(T) (STGN,fresh(T), solid line) and scaled ones (STGSC,RTC(T), dashed line), respectively (Fig. 2a, b). These two types of STG(T) together are hereinafter referred to as STGs for simplicity unless otherwise specified. The corresponding T50 and VFRavg are shown in Fig. 2c, and the sampling periods for FIGAERO-CIMS thermograms are highlighted with colored areas in Fig. 1. For each SOA system of interest, similar mass concentration of organic material after size selection was ensured for both dry and high-RH conditions so that the volatility distribution of compounds in the condensed phase was not significantly affected. For the α-pinene case, the mass concentrations of organic material after size selection under dry and high-RH conditions were 4.47 and 5.31 µg cm-3, respectively. For the SQTmix case, the corresponding values were 0.97 and 1.39 µg cm-3 under dry and high-RH conditions.

Compared to the STGs of the fresh samples, the STGs of the RTC samples shifted to higher Tdesorp values with increases in T50, regardless of the RH conditions. When examining the particle desorption profiles (i.e., the STGs), we note that the removal of compounds which were thermally desorbed below 120 ∘C and the corresponding changes in T50 were more pronounced between the fresh and RTC samples at high RH as compared to those under dry conditions. Such a difference in the changes of STGs between two RH conditions agrees with our observation of faster particle evaporation rates in the presence of water (see Fig. 1).

Under dry conditions, a larger fraction of LVOCs and ELVOCs (collectively (E)LVOCs) contributed to the STGs of the SQTmix SOA particles, with higher values of T50 when compared to the α-pinene particles (Fig. 2a, c). Consistent with the changes in VFRavg under dry conditions, relatively less increase in T50 and decrease in the STGs were observed in the SQTmix SOA particles as well. On the other hand, similar STGs were observed for the fresh samples at high RH, regardless of SOA particle type. According to the evaporation model simulations described in a previous study using a similar measurement setup (Li et al., 2019), a majority of IVOCs and SVOCs is expected to evaporate rapidly from fresh particles during the first 8–30 min at high RH. It should be noted that during the same evaporation timescale (≤ 0.5 h), the evaporation of (E)LVOCs is expected to be negligible. Therefore, the VFRavg (tR=0.25 h) is approximately determined by the ratio of (IVOC + SVOC) / (LVOC + ELVOC) in the initial particles. As the FIGAERO sampling periods last for 30 min, it follows that under high-RH conditions, the fresh particles lost a significant fraction of the initially present IVOCs and SVOCs during sample collection. Thus, the similarity in STGs between the α-pinene and SQTmix SOA particles suggests that the (E)LVOC fraction in both SOA types had a similar volatility distribution and/or thermal desorption behavior. Note that this does not mean that the same types of compounds were present in the two different SOA types. For the same reason, the difference in T50 between the two different types of fresh particles was less noticeable than the difference in VFRavg at high RH (Fig. 2c, solid circles).

In total, five sample factors (AF1–AF5, colored) were identified for the α-pinene SOA particles as shown in Fig. 3. For AF1–AF4, average MW increased with higher T50 (i.e., lower volatility). While these factors were dominated by compounds with C≤10, as expected for a precursor composition of C10H16, additional amounts of compounds with C>10 (i.e., dimers or oligomers) contributed to the total signal of AF3 and especially to that of AF4 (see also Fig. S10a). With increasing T50 values, factors had longer carbon chain lengths and higher oxygen contents, as indicated by their average molecular composition. There was no clear association between OSc and T50 for factors AF1–AF4, since the increase in carbon chain lengths was counterbalanced by the simultaneous addition of oxygen and hydrogen numbers. Therefore, the decrease in volatility of type V factors was mainly driven by the increase in average MW.

For AF5, its bulk properties and composition distribution (Figs. 3 and S10a) were closest to those of AF2 and AF3, with compounds with MW < 200 Da dominating their factor mass spectra. However, the thermal desorption behavior of AF5 was completely different, with almost all of its signal occurring at Tdesorp>100 ∘C and a continuous increase with Tdesorp until the soak period started. Many of the compounds assigned to AF5 also showed contributions to other factors at lower Tdesorp values. It is very unlikely that all these were isomeric compounds spanning 5 or more orders of magnitude in C* between the isomeric forms. It is much more probable that those compounds with small MW in AF5 were decomposition products of thermally unstable compounds with larger MW and lower volatility (D'Ambro et al., 2018; Schobesberger et al., 2018; Yang et al., 2021).

In a similar way as for the α-pinene SOA particles, five sample factors (i.e., SF1–SF5, colored) were identified for the SQTmix SOA particles, as shown in Fig. 4. For SF1–SF4, lower volatilities characterized by higher T50 values again correlated with increasing average MW but not with average OSc. Furthermore, these factors mostly comprised compounds with C≤15 (Fig. S10b), as expected for a precursor composition of C15H24. Due to the prevalence of acyclic structures in the C15 carbon skeletons of both farnesene and bisabolene (in particular exocyclic double bonds), the investigated SQTmix is more prone to undergo fragmentation, compared with those sesquiterpenes dominated by cyclic structures (e.g., β-caryophyllene) (Faiola et al., 2019). As these smaller fragments can undergo oligomerization reaction, compounds with C < 15 can also be oligomers (e.g., a C14 compound as combination of two C7 fragments). However, elucidating the detailed formation mechanisms of the observed compounds in SQTmix SOA particles goes beyond the scope of this study.

Like the AF5 factor in the α-pinene SOA case, the SF5 factor in the SQTmix SOA case contained mainly small compounds with MW < 200 Da despite displaying a continuous increase in signals at temperature above 100 ∘C (Fig. 4). This, again, suggests that thermal decomposition was the main source process when compounds of SF5 were being desorbed from the FIGAERO filter. Consistently, the compositional profile of SF5 was also dominated by compounds with small carbon numbers (Fig. S10b).

Consistent with the small change in VFR (< 12 % in volume), the particle composition in the dry SQTmix SOA particles barely changed (Fig. 5, red colors), with negligible shifts only in the NCR of SF1. As seen in Fig. 5, for the factors SF1, SF2, and SF4, the NCR decreased with decreasing VFR, implying the contribution of evaporation to the material loss. At high RH, SF1 and SF4 were no longer present after isothermal evaporation in the RTC.

As the range of C* assigned to the characteristic Tdesorp of SF1 was high enough to enable significant evaporation during the experimental timescale of up to 4.25 h and its NCR exhibited a decreasing trend with evolving evaporation, we can conclude that the decrease in NCR of SF1 was primarily driven by evaporation. In this case, the particulate water mainly accelerated the evaporation as an effective plasticizer. The decrease in NCR for SF2 and SF4, which have volatilities in the LVOC and ELVOC range respectively, was even stronger than that of SF1 at high RH. This was surprising as compounds in that volatility range are not expected to evaporate significantly from particles within 4.5 h at room temperature (Li et al., 2019). Hence, this observation indicates that in addition to evaporation, there was another loss mechanism (i.e., aqueous-phase process) driving the evolution of SF2 and SF4 under high-RH conditions.

When investigating the factors SF3 and SF5, changes in their NCR were negligible under dry conditions, but significant increases in their NCR were seen at high RH (Fig. 5). At the same time, we can see that both of these factors accounted for substantial amounts of the total particle composition at high RH (Fig. 4). This clearly indicates that compounds in SF3 or SF5 were not only retained in particle phase due to their low C* values in the range of (E)LVOC, but also formed in the particle phase at high RH. These processes must be relatively fast as the changes in abundance and NCR were already clear at the fresh stage (i.e., within 0.25 h).

Except for SF1, all factors showed a distinct shift to higher values of characteristic Tdesorp under high-RH conditions as compared to dry conditions. This also indicates that the presence of water content has a more complex impact on the particle composition than simply enhancing the isothermal evaporation of volatile compounds. The correlations induced by the aqueous-phase processes are more important than the grouping solely by volatility class. That is, compounds with a wider range of volatilities may be grouped into a factor if they are produced by the same chemical process. We provide additional discussion about the possible reasons for the changes of the factor thermogram shapes in the Supplement (see Sect. S1.2.3).

We will further elaborate on the possible reasons for these observed changes in NCR together with those described in the next section for α-pinene SOA particles in Sect. 3.3.3.

The response of the STG to isothermal evaporation and humidification appeared to be very similar for the α-pinene and SQTmix SOA particles (Fig. 2a, b). The investigation of the NCR values of PMF factors revealed that, while the overall behavior was indeed similar, there were also some distinct differences in the chemical composition between these two types of SOA particles.

As expected from the isothermal evaporation measurements and the comparison of the STGs before and after isothermal evaporation in the RTC, the α-pinene SOA particles showed very little change for the NCR under dry conditions (Fig. 6, red colors). Under high-RH conditions, AF1, AF2, and AF4 exhibited lower NCR values (NCR < 1) compared to the dry conditions (Fig. 6). However, a continuous reduction in NCR with decreasing VFR (to the point that no contribution of the factor is detectable) was only observed for AF1. Similar to the case of SF1, we concluded that the evolution of AF1 was primarily driven by the evaporation process controlled by its average C*, which lies in the volatility range between SVOC and LVOC.

The evolution of NCR with decreasing VFR was more complex for AF2 and AF4 as compared with that for AF1: their NCR values did not decrease with decreasing VFR but instead showed an increase with decreasing VFR at high RH. These observations imply that the aqueous-phase chemical transformations were the dominant processes affecting the evolution of AF2 and AF4 at high RH instead of simple evaporation. Such chemical transformations could also cause the increases in the characteristic Tdesorp and the factor thermogram width observed at high RH (Figs. 3a and 6a), in particular for AF2 with its T50 increasing from 105 to 135 ∘C.

AF3 exhibited an NCR > 1 in the fresh case under high-RH conditions, which means additional amounts of compounds grouped into that factor were formed in the presence of an aqueous phase in the particles. Note that many of the ions grouped into AF3 also showed an increase in the absolute measured signal under high-RH conditions after accounting for the different amount of collected sample mass on the filter. With longer isothermal evaporation time, NCR decreased for AF3, which means that some of the compounds grouped into AF3 must have evaporated from the particles or continued to react to form different products grouped into other factors. The change of the factor thermogram shape (i.e. loss of compounds with higher C* and lower Tdesorp) in Fig. 3 together with a minor shift in the characteristic Tdesorp in Fig. 6 suggests that the removal due to evaporation is the more likely explanation. Hence, the evolution of NCR of AF3 at high RH suggests complex behavior including the formation of compounds at the particle phase but also the loss of some compounds mainly by evaporation.

Negligible changes in NCR of AF5 alone indicate minor changes in composition during evaporation under dry or high-RH conditions. In addition, when considering that AF5 is (mainly) in the ELVOC range (see Fig. 3), the isothermal evaporation of compounds should not be significant in the experimental timescale of up to 4.5 h (Li et al., 2019). But when investigating the factor thermograms (Fig. 3) in detail, the changes in the shape of the factor thermogram and Tdesop (Fig. 3) together imply that, apart from evaporation, water-driven aqueous-phase processes also affected at least some of the compounds with extremely low C*, which were grouped into AF5. Although both AF5 and SF5 were dominated by products of thermal decomposition, it does not indicate that their compositions were similar. While the mass spectra of AF5 was dominated by ions with Cnum from 7 to 10, major ions in the mass spectra of SF5 tended to have Cnum of 6 or below (Fig. S10). As these two factors originated from two different SOA systems, it is highly possible that they can behave differently against particulate water. It is also important to remember in this context that the products of any decomposition process may be similar or even identical, but they may stem from completely different parent compounds. In particular, very small fragments (e.g., oxalic acid or acetic acid) carry very little information about the original molecule they came from.

Overall, particulate water not only accelerates the evaporation of sample factors by reducing bulk diffusion limitations, but also alters the chemical composition of particles by inducing chemical aqueous-phase processes (e.g., hydrolysis or oligomerization). Accelerated evaporation primarily driven by the water plasticizing effect was observed for those sample factors with the smallest average MW and the highest volatility in both SOA systems (i.e., α-pinene, AF1; and SQTmix, SF1). On the other hand, changes in the NCR together with changes in the absolute abundance and/or the characteristic Tdesorp for the other sample factors very likely suggest the presence of aqueous-phase processes that generally modify the composition and volatility of the (remaining) SOA particles.

The factors affected by chemical aqueous-phase processes can be classified as (i) “educt” factors with NCR <  1 and (ii) “product” factors with NCR > 1 under the same conditions. Educt factors contain water-labile compounds which are stable under dry conditions but undergo chemical reactions in the presence of water. Likely aqueous-phase reactions are the fragmentation (hydrolysis) of organic (hydro)peroxides (Krapf et al., 2016; Zhao et al., 2018; Qiu et al., 2019) or accretion reactions. Examples for these educt factors were SF2, SF4, AF2, and AF4. All these factors exhibited NCR values clearly < 1, while their volatilities were in the (E)LVOC range, which makes a substantial isothermal evaporation within 0.25 h very unlikely.

The products of these aqueous-phase reactions will evaporate from the particle phase if their volatility is high enough (e.g., small fragments from fragmentation reactions). Products with sufficiently low volatility will remain in the particle phase and contribute to the product factors. Such compounds with sufficiently low volatility may be the larger fragments of fragmentation reactions, but the majority is likely formed from accretion reactions such as (i) non-oxidative reactions involving two or more carbonyls (i.e., (hemi)acetal formation, aldol condensation, and esterification) or (ii) reactions incorporating carbonyls and organic hydroperoxides (i.e., peroxy(hemi)acetal formation) (Kroll and Seinfeld, 2008; Herrmann et al., 2015). The predominant non-oxidative nature of these reactions is dictated by the fact that the average OSc of the particles does not increase under high-RH conditions.

The product factors for the SQTmix SOA particles (SF3 and SF5) were also identifiable by the fact that they have almost no contribution to the total signal under dry conditions. The comparable product factor for the α-pinene SOA particles (AF3) already contributed to the particles under dry conditions and then showed an increase in contribution under high-RH conditions. This behavior is probably linked to the SOA production inside the OFR which was at ∼ 40 % RH. For α-pinene SOA, compounds grouped into AF3 could be already produced inside the OFR either in the gas phase or by the particle phase processes. The absence or very small contribution of SF3 or SF5 under dry conditions indicates that the processes leading to their formation were too slow to produce significant amounts during the short residence time prior to the particle size selection.

Another difference between the two SOA types lies in the evolution of the educt and product factors in the RTC under high-RH conditions. For the SQTmix SOA particles, the evolution of the NCR values of all factors was monotonic (i.e., either increasing or decreasing with decreasing VFR). This may indicate that the underlying dominant process is either a removal or a production process for each factor. It should be noted that multiple loss and production processes may coexist for a factor, especially at high RH where aqueous-phase processes may play a role. For instance, the removal of compounds grouped into the educt factor AF2 or AF4 via chemical reactions was dominant over any production process. But with increasing isothermal evaporation time at high RH, the balance between these processes shifted slightly, leading to a small increase in the NCR. The balance between the removal and production of compounds may vary over time. This is probably the cause of the complex behavior of NCR values for AF2, AF3, and AF4 and may be coupled to the observed changes in the factor thermogram shapes for these factors.

Although there are multiple studies of isothermal evaporation of α-pinene SOA particles, very few studies provide molecular information that is comparable to our approach. D'Ambro et al. (2018) conducted FIGAERO-CIMS measurements of particles that evaporated on the filter after collection. Although the α-pinene SOA particles in the study may not be directly comparable to the particles in our study, some of their findings share similarities with the interpretation of our PMF factors. For each ion, they explain the observed isothermal evaporation behavior with a model containing three components with different apparent volatility: (i) free monomers that evaporate from particles according to their C* values, (ii) ELVOC compounds that do not evaporate from the particles at room temperature but decompose upon heating to be detected as the single ion, and (iii) reversible oligomers that decompose into the corresponding free monomers with time or heat. In our data set, many individual ions show contributions from multiple factors. AF1 and SF1 are predominantly containing compounds that behave like free monomers. AF5 and SF5 are mostly ELVOC compounds that are detected as thermal decomposition products. The behavior described for reversible oligomers is in line with the complex behavior of the PMF factors which we associate with aqueous-phase processes. As D'Ambro et al. (2018) only applied their model investigation to particle evaporation at 50 % RH and above, it is impossible to determine whether the particle phase processes affecting the reversible oligomers are linked to the presence of particulate water. Note that the approach of D'Ambro et al. (2018) deploys a ion-by-ion model fitting, while our PMF analysis inspects the behavior of all ions in the data set at once.