The most abundant component of modern jet lubrication oils is a mixture of synthetic ester blends (Winder and Balouet, 2002; Craig and Barth, 1999). Additives are also present in jet lubrication oils, which improve their resistance to degradation (Duong et al., 2018), although these vary by manufacturer and specific application. AeroShell Turbine Oil 390 was selected for this study and contains a mixture of synthetic esters and additives. It has a low toxicity and a low viscosity (12.9 mm2s-1 at 40 ∘C), which allows it to be atomized safely and effectively at low pressures.

The experimental setup used to perform measurements is illustrated in Fig. 2. This can be decomposed into five components: the inlet system, the droplet generation system (Sect. 3.3), the aerosol chamber, the size distribution measurement system and the expansion system (Sect. 3.4). The flow of aerosol through each of the five components was driven by a Titan Precision TT-25L-SILENT air compressor and a series of vacuum pumps. Before being exposed to the sample aerosol, air was first cleaned by passing it through an activated carbon filter (Pall Carbon Capsule 12011) to remove residual organic material and then through a Whatman high-efficiency particulate air (HEPA) filter to remove ambient aerosol.

Lubrication oil droplets were generated using a Collison three-jet nebulizer (May, 1973). To minimize droplet concentrations, the nebulizer was run at its lowest operable pressure of 20 psig corresponding to a flow rate of approximately 5 Lmin-1. Previous measurements with a similar jet lubrication oil indicated that, under these conditions, particles were generated with an integrated number concentration exceeding 106 cm-3. The condensation particle counter (CPC), TSI model 3750 and optical particle counter (OPC), Palas Welas 2500, have integrated number concentration upper limits of 105 and 4000 cm-3, respectively. For this reason, several measures were taken to reduce the particle number concentration. Firstly, the Collison outlet was fed directly into a 0.9 m3 aerosol chamber, whereupon aerosol was diluted with filtered air. Secondly, the compressed air inlet for the Collison three-jet nebulizer was opened for a maximum duration of 1 s. These were denoted as “priming” events and were typically performed every 100 min to maintain droplet number concentrations at least 1 order of magnitude greater than background number concentrations. (See Sect. S1 in the Supplement for additional information.)

The flow of compressed air into the aerosol chamber was regulated at the inlet using a mass flow controller. To maintain the aerosol chamber at ambient pressure, the sum of flow rates in and out of the chamber was matched, except during “priming” events where the volume of aerosol passed into the chamber was negligible. The first aerosol chamber outlet was connected directly to the particle-sizing apparatus, which comprised an electrical aerosol neutralizer (TSI model 1090) and a Scanning Mobility Particle Sizer (SMPS, TSI model 3938) in series. The particle-sizing apparatus allowed for particle size distribution within the aerosol chamber to be measured continuously when interfaced with the Portable Ice Nucleation Experiment (PINE).

The second aerosol chamber outlet was connected directly to the Cambustion Aerosol Aerodynamic Classifier (AAC). The AAC was used to select particles with a narrow range of aerodynamic diameters from the otherwise polydisperse particle size distribution in the aerosol chamber. The particle size distributions measured in the aerosol chamber and at the AAC outlet are related by the AAC transfer function, which describes the fraction of particles that successfully navigate the AAC as a function of particle diameter. Therefore, if both the AAC transfer function and the particle size distribution within the aerosol chamber are known, their product provides an indirect estimate of the particle size distribution at the AAC outlet (PINE inlet).

To characterize the transfer function, size distributions were recorded with and without the AAC present. During these separate measurements, an electrical mobility diameter (dm) set point of 100 nm was selected using the AAC. To achieve this, the instrument was set to perform an internal numerical conversion between aerodynamic and mobility diameters given the density and shape factor (ρ = 924 kgm-3 at 15 ∘C, χ = 1) of the oil droplets (Johnson et al., 2018). This resulted in an experimental transfer function with a geometric mean diameter (GMD) of dm = (100.9 ± 0.6) nm and a geometric standard deviation (GSD) of 1.08 ± 0.01 (see Sect. S2 in the Supplement). The experimental transfer function was narrower than the theoretical AAC transfer function derived with inlet and size-dependent particle losses and resembled the idealized theoretical transfer function (Johnson et al., 2018). The narrowness of the experimental transfer function was attributed to the low vapour pressure, well-defined sphericity and density of the oil, which allowed for the conversion between dm and da to be made with high confidence (Johnson et al., 2018).

During operation, the AAC outlet was connected to an inline aerosol flowmeter (AF10, Cambustion, UK). This was used to determine the flow rate upstream of the PINE system without interfering with the aerosol sample. Aerosol was then passed through a dual-membrane diffusion dryer to reduce its dew point and limit frost formation on the PINE expansion chamber walls (Möhler et al., 2021).

In terms of the expansion measurements, the PINE instrument comprises three main stages: inlet (diffusion dryers), expansion chamber and OPC (Möhler et al., 2021). The expansion chamber has a capacity of 7 L and was cooled using a Chiller Lauda (RP855) cooler. The gas temperature within the chamber was measured using three thermocouples which are distributed vertically along the long axis, each of which was calibrated to a reference sensor with an accuracy of ± 0.1 ∘K (Möhler et al., 2021). The outlet flow passed through an OPC, which counted particles within a nominal diameter range of 0.6–40 µm. Each measurement using the PINE chamber comprised three modes: flush, expansion and refill (Möhler et al., 2021). In flush mode, approximately 21 L of aerosol flow was passed through the 7 L expansion chamber. This was to ensure that the aerosol equilibrated with the chamber walls, so that, immediately before an expansion, sample aerosol had a water partial pressure equal to the vapour pressure above ice, pw=pice (100 % RHi). During the expansion mode, the inlet valve was closed and the contents of the chamber were pumped out (via the OPC), resulting in adiabatic cooling and an increase in the sample relative humidity. Depending on the temperature of the chamber, water droplets and/or ice crystals were detected at the OPC as a function of the temperature and pressure in the PINE chamber (see Fig. 3). Note that, because the PINE chamber and OPC were connected by a short (0.2 m) length of tubing, combining OPC (particle counts) and PINE (temperature or pressure) data based on a common measurement time would introduce a non-negligible time delay due to particle travel. This time delay was factored into the calculations of onset positions, resulting in a decrease in onset relative humidity values by 1 %.

Measurements using the lubrication oil were performed using the experimental setup in Fig. 2. The aerosol chamber was first “primed”, generating average integrated chamber aerosol concentrations on the order of 1000 cm-3. During these measurements, the PINE chamber was cooled to initial chamber temperatures in the range 230 K < Ti < 250 K. This allowed for condensation and freezing events to be investigated in the temperature range 225–245 K, in line with the atmospherically relevant temperatures presented in Fig. 1. A continuous flow of aerosol was passed through the PINE system, with average integrated aerosol concentrations on the order of 100 cm-3 exiting the AAC.

First, conditions required to activate ultrafine lubrication oil droplets to form water droplets were investigated. For each expansion measurement, change-point analysis was used to identify the chamber pressure at which the OPC recorded a significant change in the particle count. Specifically, the point at which threshold values were met by both the particle concentration and its time derivative was noted as the onset position. Combining the onset pressure with the pressure and temperature before an expansion allowed for the onset temperature, To, to be determined (see Sect. S3 in the Supplement). The Si at the onset (Si,o) could also be determined using these pressure and temperature values, enabling the phase space in Fig. 1 to be populated with onset positions for each expansion. As mentioned in Sect. 2, the calculation of Si is sensitive to the parametrization of pliq. Two parametrizations were compared, and it was found that the more recent parametrization proposed by Nachbar et al. (2019) deviates from the parametrization proposed by Murphy and Koop (2005)) by < 1 % over the range of onset temperatures explored within this work (T > 225.3 K). The authors have therefore chosen to parametrize pliq according to Murphy and Koop.

Each of the marker types in Fig. 4a corresponds to a set of repeat measurements with the PINE chamber at a similar initial temperature (with an average temperature of the set Ti). Uncertainties in the initial pressure values were estimated using the standard deviation of the final 300 equilibrated pressure values recorded prior to an expansion. Uncertainties in the onset pressure values were limited by the instrumental resolution (1 Hz) and determined conservatively as half the average pressure difference between adjacent values during the expansion. Pressure and temperature (± 0.1 K) uncertainties were used to derive the error bars on the onset saturation ratio (ΔSi,o). For a given Ti, these pressure uncertainties lead to onset positions being distributed along the corresponding dry adiabat, accounting for the trend shown in Fig. 4a.

Repeat experiments performed with similar initial chamber (and therefore onset) temperatures were averaged to produce Fig. 4b. For comparison, onset positions for citric acid aerosol have been shown in Fig. 4b (see Sect. S4). These were obtained using the PINE chamber, and the experimental data were analysed according to the relations outlined in Sect. S3. The onset positions for citric acid aerosol follow the expected behaviour for a hygroscopic aerosol: they are positioned along the water saturation line (Wilson et al., 2012; Murray et al., 2010a) until the bifurcation at ∼ 234 K, after which they extend along the Koop line (Koop et al., 2000). This agreement affords confidence in the onset positions for the lubrication oil droplets obtained in this work, since these were investigated using similar methods.

The onset positions of lubrication oil droplets lie above the water saturation line for all the temperatures investigated (225–245 K). These onset positions are consistent with the Kelvin equation, which represents the limiting activation behaviour of wettable, insoluble aerosol particles, which are characterized by a hygroscopicity parameter κ=0, in accordance with κ-Köhler theory (Petters and Kreidenweis, 2007) (see Sect. S5). This result was anticipated on account of the lubrication oil's negligible water solubility (Sullivan et al., 2009). To place the behaviour of lubrication oil in the context of contrail formation, a comparative study on the onset behaviour of aircraft-generated nvPM is required.

Despite the wealth of literature concerning the onset behaviour of nvPM (Mahrt et al., 2018; Koehler et al., 2009; Möhler et al., 2005; Gao et al., 2022b; Gysel et al., 2003), there is a paucity of information concerning the onset behaviour of aircraft-generated nvPM or representative surrogates (Marcolli et al., 2021) and their dependence on temperature. The example nvPM onset positions presented in Fig. 4b are taken from recent experiments performed using size-selected miniCAST-generated soot with a modal diameter of 200 nm (Gao and Kanji, 2022a). This study sits within a body of research that provides evidence in support of the pore condensation and freezing (PCF) mechanism for large (100–400 nm) porous soot aggregates (Marcolli et al., 2021; David et al., 2019). During PCF, water is collected within mesopores (2–50 nm) by capillary condensation under subsaturated conditions, after which it freezes. To our knowledge, PCF has not been measured for soot particles with sizes that are representative of aviation nvPM, and it is challenging to extrapolate the implications of PCF for representatively sized nvPM, particularly given additional complications associated with onset sensitivity to internal and external mixing of semi-volatile material (Gao and Kanji, 2022; Gao et al., 2022b, a). Given the onset positions for lubrication oil droplets follow the extremal curve for an insoluble aerosol and based on the available literature referenced above, it is reasonable to conclude that lubrication oil droplets will activate at supersaturations no lower than those required to activate aircraft-generated nvPM. In other words, aircraft-generated nvPM particles are likely to be more effective CN than ultrafine lubrication oil droplets.

During contrail mixing, modest deviations from TC result in peak supersaturations of ∼ 10 % (Bier et al., 2022; Kärcher et al., 2015), depending on the properties of nascent aerosol in the plume. According to Fig. 4b., these conditions would be sufficient to activate a proportion of both nvPM particles and lubrication oil droplets. For contrails forming below threshold conditions (TA ≪ TC), it is therefore feasible that lubrication oil droplets could competitively deplete plume supersaturations during contrail mixing under soot-poor conditions. However, under near-threshold conditions (TA ∼ TC), where only infinitesimal supersaturations are achieved, it is likely that nvPM will preferentially activate in favour of the less active lubrication oil droplets. These implications have previously been corroborated computationally (Kärcher and Yu, 2009).

As discussed in Sect. 4.1, the activation of lubrication oil droplets is important for predicting their behaviour under contrail-forming conditions. Additionally, the freezing mechanism(s) for these droplets and the temperatures at which these mechanism(s) are operable are also important for understanding the behaviour of the resulting ice crystals under contrail-forming conditions.

To elucidate the freezing mechanism(s), particle size distributions were investigated as a function of temperature through individual expansions; these are shown in Sect. S6. Four distributions are displayed in Fig. 5 and serve to disambiguate the freezing mechanism(s). The expansion corresponding to To = 244.2 K produced a monomodal distribution with a GMD of (5.25 ± 0.01) µm and a GSD of 1.100 ± 0.002, resulting from the activation of lubrication oil droplets into water droplets (as described in Sect. 4.1). For the expansion with an onset temperature of To = 240.0 K, the droplet mode was again visible at a comparable GMD of (3.4 ± 0.1) µm (GSD = 1.72 ± 0.05), persisting down to a temperature of ∼ 235 K. However, when the temperature within the chamber decreased below this value, the droplet mode disappeared and a second particle mode appeared at a larger GMD of (15.4 ± 0.1) µm (GSD = 1.47 ± 0.01). Owing to the relative size difference between the two modes (Lacher et al., 2017), the latter mode can be attributed to ice crystals which form when a fraction of the activated lubrication oil droplets subsequently freeze.

During the next coolest expansion (To = 230.2 K), the particle size distribution grew to a GMD of approximately (21.6 ± 0.1) µm (GSD = 1.276 ± 0.005). The expansion at To = 224.9 K demonstrates similar behaviour with a comparable distribution width GSD = (1.27 ± 0.01), albeit with a smaller GMD of approximately (14.5 ± 0.1) µm. The size distributions for the two expansions at cooler temperatures feature the familiar ice crystal mode at larger nominal particle diameters (> 5 µm). Based on the analysis presented in Sect. 4.1, condensation must have taken place shortly after water saturation was reached. This implies that, during the two expansions performed at cooler temperatures, ice crystals nucleated from growing water droplets. This process is referred to as condensation and freezing and has been well-documented in the literature (Murray et al., 2012)

Based on the evidence in Fig. 5, the microphysical pathway to ice crystal formation has been partly justified as condensation followed by freezing. Supercooled water droplets containing aerosol particles can freeze either heterogeneously or homogeneously. Homogeneous freezing occurs when water molecules on the surface of the aerosol particle reach a critical cluster size, which depends upon the droplet volume, temperature and freezing rate. Once the critical cluster size is reached, the rate of homogeneous freezing becomes comparable to the experimental timescale and freezing occurs readily (Murray et al., 2012). The homogeneous freezing temperature for micrometre-sized droplets has been demonstrated to lie in the range (235–236) K, as verified by numerous experimental studies using a variety of techniques (Tanaka and Kimura, 2019; Zhang et al., 2012; Murray et al., 2010b; Tarn et al., 2021). Using the methods outlined in Koop and Murray (2016) and Murray et al. (2010b), the onset of homogeneous freezing under the conditions in Fig. 5b was estimated as ∼ 235 K (see Sect. S7). In contrast, for a water droplet to freeze heterogeneously, the immersed aerosol particle must have surface properties that lower the activation energy to ice nucleation (Knopf et al., 2018; Knopf and Alpert, 2023). For this reason, heterogeneous freezing temperatures are necessarily warmer than homogeneous freezing temperatures under otherwise comparable conditions (Kärcher and Lohmann, 2003).

It is important to understand the freezing mechanism adopted by contrail ice-forming particles as this influences the size distribution of ice crystals and hence the optical properties of the contrail. To demonstrate this, consider a parcel of air cooling in the atmosphere consisting of both heterogeneous and homogeneous ice-forming particles. If we assume that both particle types activate to form water droplets under similar conditions, cooling droplets that contain the heterogenous ice nuclei will freeze preferentially, as reasoned above. Computational and experimental studies on cirrus cloud formation suggest that because these freezing events deplete the supply of water vapour, they can act to delay or even inhibit homogeneous freezing in the remaining droplets (DeMott et al., 2003; Spichtinger and Gierens, 2009). As there is less competition for growth, the ice crystals will be larger than those that would have formed had all the droplets frozen homogeneously; however, because only a fraction of the droplets have frozen, there will be fewer ice crystals overall (Spichtinger and Gierens, 2009). As a consequence, the total droplet surface area is reduced, which is an important parameter controlling optical depth and thus radiative forcing (Frömming et al., 2011; Kärcher, 2016). Note that these conclusions were drawn from experimental and computational measurements concerning cirrus clouds, which have rates of Si and concentrations of ice-forming nuclei that differ by several orders of magnitude from those experienced within a contrail mixing plume. The above discussion therefore serves only to illustrate that the freezing mechanism adopted by ice-forming particles within a cooling air parcel influences its optical properties.

In Fig. 5b, freezing is observed during water droplet growth at a temperature of ∼ 235 K, which lies within the expected range for homogeneous freezing and is consistent with the theoretical value identified using Fig. S7. Therefore, it is likely that water condensed on the lubrication oil droplets freezes homogeneously rather than heterogeneously. Finally, the microphysical pathway to ice crystal formation on lubrication oil droplets has been justified as condensation followed by homogeneous freezing. This pathway is illustrated pictorially in Fig. 6.