Laboratory studies of the physiochemical properties of mineral dust require an appropriate sample material that is representative and relevant to atmospheric processes. Field studies show that most dust in the atmosphere is dominated by varying quantities of quartz, K-feldspar, plagioclase, calcite, hematite, kaolinite and the illite–smectite group of clay minerals (Kaufman et al., 2005; Kandler et al., 2007, 2009, 2011; Formenti et al., 2008; Jeong, 2008; Kaufmann et al., 2016). At a single-particle level, particle size dependence is observed with an abundance of quartz and feldspar grains in the coarse fraction and a fine fraction (<2 µm) dominated by clay minerals. Detailed studies on Asian and Saharan dust have revealed complex internal structures of individual particles in terms of mineralogy and morphology. TEM analysis of sliced particles revealed that the most common sub-5 µm particle type was clay-rich agglomerate, dominated by nano-thin platelets of illite–smectite series clay minerals (ISCM) with submicron grains of iron (hydr)oxides (goethite and hematite) commonly dispersed throughout the particles (Jeong and Nousiainen, 2014; Jeong et al., 2016).

In our laboratory study of silicate dust, clay mineral samples were chosen because they represent the most atmospherically relevant material in the size range of the LAAP-TOF transmission (0.5–2.5 µm). Samples of feldspar were compared to these clay samples because of the potential importance of K-feldspar as ice nuclei (Atkinson et al., 2013). In addition to aluminosilicates, 2.1 µm borosilicate glass spheres were used as an example of an amorphous silicate structure that contains abundant alkali and earth alkali metals that are not chemically bonded into a crystal structure. Carbon black (CB) (Elftex 124, Cabot Corp) was used as an example of a non-silicate particle that has a well-characterised molecular structure. A summary of the samples used in this study is provided in Table 1.

CB is distinct from the material commonly referred to as black carbon (BC) (Watson and Valberg, 2001; Long et al., 2013). CB has a characteristic particle morphology that consists of spherical primary particles fused into aciniform (grape-like) aggregates which cluster into larger-sized agglomerates. The primary particles are typically 10–500 nm in diameter and are composed of imperfect graphitic layers that are concentrically arranged around a growth centre (Rivin, 1986). The spaces between the graphitic layers often accommodate cations such as potassium and sodium. Elftex 124 is a flame-soot-derived CB product comprised of >95 % carbon (Wonaschütz et al., 2009).

Clay minerals have a sheet silicate structure in which silicate tetrahedra are 2-dimensionally polymerised to form structural layers that are separated by interlayer cations. The capacity of a clay mineral to accommodate interstitial cations (X-ions) in the interlayer spacing of a clay mineral is related to the layer charge of the structural layer. The structural layers of clay minerals are composed of alternating tetrahedral and octahedral sheets in which the layer charge is created by isomorphous substitution of lower valency ions. Substitution of Si4+ for Al3+ ions in the tetrahedra produces a tetrahedral charge, while substitution of Al3+ for Mg2+ in the octohedra (Y-ions) produces an octahedral charge. The composition of clay minerals is not fixed but is determined by the varying degrees of cation replacement in the structural unit and interstitial complex. For example, the nominal chemical formula of illite is stated in the form K1.5-1.0Al4[Si6.5-7.0Al1.5-1.0O20](OH)4 to express the variability in composition that can exist within the illite member of the clay mineral group.

Source clays from the Clay Minerals Society (CMS) have been the subject of chemical, physical and thermodynamic analysis and therefore very well characterised. Average elemental composition of Si, Al, Fe(II), Fe(III), Mg, Ti, Mn, P, Ca, Na, K and H2O of the bulk sample is available (Mermut and Cano, 2001) and enables the construction of the precise structural formula. The structural formulae for the source clay samples used in this study are given in a format that shows the average composition of the X-ions, Y-ions and tetrahedral layer in Table 2.

Kaolinite (samples KGa-1b and KGa-2) has a 1:1 layer structure in which each layer contains one tetrahedral layer and one octahedral layer. Kaolinite has no overall layer charge and hence a very low X-ion content. The remaining samples consist of the so-called 2:1 layer clays in which one octahedral layer is sandwiched between two tetrahedral layers. The layer charge created by this 2:1 layer structure is balanced by a certain quantity of X-ions. Illite-rich samples (IMT-2, ISCz-1) are characterised by very high quantities of K+, whereas the montmorillonite clays (STx-1b, SWy-3) are characterised by relatively high levels of Na+ and Ca2+.

Naturally occurring rock is rarely monomineralic. Mineralogical impurities in source clays have been previously quantified by X-ray diffraction analysis (Table 3) for the bulk sample (Chipera and Bish, 2001) and <2 µm fraction (Vogt et al., 2002). For example, in natural environments montmorillonite is diagenetically altered to illite in the presence of K-feldspar (Garrels, 1984) so that clays often consist of fine interlayers of these two clay minerals. Sample ISCz-1 is an example of a clay that consists of microscopic interlayers of illite and smectite clay minerals (ISCM). Recently, Broadley et al. (2012) suggested illite NX (B+M Nottenkamper, Munich, Germany) as a suitable representation of ambient mineral dust sampled at remote locations and it has been used in numerous ice nucleation studies (Hiranuma et al., 2015). Illite NX is a clay-rich nanopowder that contains significant mineralogical impurities. XRD analysis has shown significant variation in the impurity content that may represent significant batch differences in mineralogy (Table 4).

In framework silicates such as feldspar, the silicate structure is formed from 3-dimensionally polymerised silicate tetrahedral. In contrast to the interlayer cation structure of clay minerals, a low-valence cation is held in an interstitial cavity that balances the charge deficit created by substitution of Al3+ for Si4+ in the tetrahedral. Unlike clay minerals, the interstitial cations are fixed and not exchangeable during normal diagenetic processes. Most naturally occurring feldspars are not homogenous but contain separate potassium-rich and sodium-rich phases. Specimens of orthoclase (K-feldspar) and plagioclase were purchased from Geo Supplies Ltd, and analysed by XRF analysis to determine the average elemental composition. The average structural formula of the samples was then calculated from the relative proportions of the metal oxides (Table 5).

Mineral dust aerosol was sampled by the LAAP-TOF after dry dispersion of powder in a home-made dust tower and transmission of the extracted aerosol through a centrifugal particle mass analyser (CPMA) (Cambustion, Ltd) (Olfert and Collings, 2005). Source clays, illite NX, borosilicate glass beads and CB were supplied in powder form or loose aggregate that was easily broken down into powder by gentle abrasion with a pestle and mortar. The feldspar samples were supplied as a large crystal that had to be reduced to powder in a timor mill prior to injection into the dust tower.

Powdered mineral dust was loaded into a modified sample vial and injected into the dust tower inlet using a short burst of compressed air from an air duster can. The mixing and suspension of mineral dust particles was achieved by turbulent flow, controlled with an adjustable flow of dried, filtered and oil-free compressed air (produced on site using a compressor) introduced into the bottom of the tower (Fig. 1). Aerosol was drawn out of the dust tower through the CPMA using a pumped sampling line throttled to a suitable flow rate (∼1.6 Lmin-1) using a needle valve. The flow was divided between the pumped line and the LAAP-TOF inlet (sampling at 0.078 Lmin-1) using a “Y piece”.

The experiment was operated in two modes. In polydisperse mode, the CPMA was inactive, so the mass distribution of the aerosol analysed by the LAAP-TOF was determined by the transmission efficiency of the aerodynamic lens. In monodisperse mode, the CPMA was used in static mode so that only the selected particle mass was transferred to the LAAP-TOF.

The LAAP-TOF used in this study is a modified version of the commercially available single-particle mass spectrometer manufactured by AeroMegt (GmbH) that features a modified optical detection system as described in detail by Marsden et al. (2016). Briefly, particle-laden air enters the instrument via an aerodynamic lens inlet (Liu et al., 1995) which produces a narrow but divergent particle beam along the instrument axis. A pulsed excimer laser interacts with a particle in the low-pressure source region towards the back of the instrument (Fig. 2). An optical particle detection stage located within the source region is used for temporal alignment of the excimer laser pulse with the presence of a particle in the ion source. The option to measure aerodynamic particle size using an additional optical particle detection stage located upstream of the instrument axis was not used in this study.

The version of the instrument used in this study features an aerodynamic lens inlet (model LP2.5 Aeromegt GmbH) for transmission of particles approximately 0.07–2.5 µm in diameter that is similar in design to the high-pressure lens (HPL) described by Williams et al. (2013). Divergence of the particle beam is size and shape dependent (Jayne et al., 2000), resulting in a morphological dependence on the fraction of particles that reaches the ionisation region (Huffman et al., 2005), an effect that is particularly important for measurements of irregularly shaped platy mineral in polydisperse dust.

Laser desorption ionisation takes place in a high vacuum (<1 × 10-6 mbar) region some 230 mm from the aerodynamic lens exit, where particles interact with a 8 ns pulse of 193 nm radiation that is fired co-axial but counter-propagate to the particle beam. The ablation laser, an ArF excimer (model EX5, GAM Laser Inc.), can be requested to produce 2–12 mJ per pulse via a setting in the software. The peak laser power actually encountered by a particle is unknown as the particle will transect the laser pulse at an unknown position in its profile due to particle divergence and temporal alignment. Consequently, particle-to-particle variations in the peak power and effective pulse duration can reduce the reproducibility of the measurement. A previous study with this instrument configuration has shown a sampling efficiency for ambient mineral dust of approximately 0.01 with a particle-size-dependent optical detection bias that favours the detection of particles with a mode of around 1.5 µm in diameter (Marsden et al., 2016). Sampling efficiency is important in SPMS as it influences the particle-counting statistics of an aerosol population and can introduce a bias of compositional measurement towards a particular particle type (Zelenyuk et al., 2009).

The TOF-MS used in the LAAP-TOF is a bipolar reflectron TOF analyser (BTOF, Tofwerks AG) for the simultaneous measurement of positive and negative ions. Ion arrival times at the multichannel plate (MCP) detectors, one for each ion mode, are recorded by a dual channel 14bit analogue to digital converter (model ADQ214, SP Devices) with a bin width of 2.5 ns. The ADQ clock is triggered by a signal that is synchronous with the firing of the ablation laser so that each spectrum represents the ions formed by a single-particle ablation event. The configuration of the extraction optics in the LAAP-TOF is unusual in that the first extraction lenses of the LAAP-TOF ion optics are grounded so that the evolution of the ion plume takes place in a field-free region and only ions with the correct ion trajectories are extracted for analysis.

The aim of the peak analysis is to evaluate how composition and crystal structure of the particles affect ion arrival times at the multichannel plate (MCP) detector by comparing the spectral characteristics of the samples of mineral dusts of known mineralogy. TOF-MS analysers focus a certain elemental or molecular ion species into a discrete packet of ions with arrival times at the detector that are represented by a Gaussian probability density function (PDF), which is represented as a spectral peak in the data. The average ion arrival time (T) of the ion packet at the detector is described by the peak centroid, while the peak width (FWHM) describes the distribution around this average.

Peak centroid and peak width were extracted from the raw spectra using a peak-fitting algorithm. Peak fitting within the spectra was problematic because of spectrum-to-spectrum variations in peak position, peak shape and interference from neighbouring peaks. To reduce the influence of these effects, a multipeak-fitting procedure (Multipeak fit v2, igor v6.36) was performed on a portion of the mass spectrum which contained the peak of interest. Peak-fitting parameters were set so that interfering signals were identified as separate peaks without comprising the integrity of the peak of interest. In most cases, the peak of interest was a well-defined Gaussian shape that was sufficiently separated from its neighbour, such as the example of SiO3- molecular ion (m/z-76) in Fig. 3a. Occasionally, especially in illite-rich clay mineral samples, the molecular ion peak was much broader (Fig. 3b) therefore a number of data smooths were required to ensure satisfactory peak definition was obtained. A third peak morphology, featuring multiple modes was also encountered in illite-rich clays samples that required careful consideration. Using a suitable peak-fitting parameters (Number of Smooths 5, Number Fraction 0.1), the multimodal peak in Fig. 3c was resolved into two peaks whose peak centres were approximately 1 Da (50 ns) apart. Note that further resolution of this multimodal peak using a fewer data smooths resulted in peaks that were separated by <0.5 Da (20 ns) (Fig. 3d), which is an improbable solution in data that primary consists of singly charged ions. Consequently, peak-fitting parameters were chosen that ensured peak centroid spacing of integer values (in mass) were produced from all fitting. The origin of the multimodal peak morphology will be discussed in more detail in the results section.

Prior to analysing the mineral dust samples, the ion optics were tuned for maximum resolution and symmetrical peak shape across the mass range for positive and negative ions using 700 nm diameter PSL particles. The effect of particle type on ion arrival times was initially evaluated by comparing mass-scale calibrations obtained from selected samples. Calibration coefficients were calculated for each sample type by fitting the first order approximation of the time of flight equation (Eq. 1) to the averaged peak centroid of three ion species. For the silicate, the same ion species were chosen in positive and negative ion modes for each sample. Changes in average arrival times of a certain ion species (i), apparent as a peak centroid shift (ΔT(i)) with respect to previously defined position, were examined in more detail by comparing the peak centroid with a nominal mass-scale calibration and comparing ΔT(i) with other ion species from within the spectrum (i.e. from the same ionisation event). 1T=a+bM

Spectral resolution of the peaks in positive and negative spectra for PSL were typically 300 and 600 (FWHM at m/z 36) respectively, with unit mass resolution easily achieved over the mass range 10–100 Da. The optimum tune settings for the positive ion optics differed somewhat from those of the negative ion optics despite their identical geometry, indicating significant differences in the translational energies of the positive and negative ions produced in the same particle ionisation event.

Typical mass spectra of feldspar mineral dust (Fig. 4a) show elemental ions of alkali metals in positive ion mode and molecular fragment ions of silicate at m/z -60 (SiO2) and m/z -76 (SiO3) in negative ion mode, which is in agreement with the mineral dust spectra observed by Gallavardin et al., 2008a, who used a similar 193 nm excimer laser at approximately 109 Wcm-2. Detector saturation was apparent on several positive ion peaks when Channel 1 signal amplifier line to the A/D was used (Fig. 4a, top). To avoid peak saturation, the Channel 2 signal amplifier with a greater attenuation was used for the subsequent peak analysis. The effect of changing the signal amplifier on the spectra is demonstrated in the typical mass spectra examples in Fig. 4. Detector saturation occurs at 1200 mV signal strength for the peaks at m/z 27, 39 and 44 with the signal Channel 1. Saturation is avoided with signal Channel 2 but comes at the expense of poor signal-to-noise ratio of ion species such as C+ (m/z 12) and CaO+ (m/z 56). Note the relative signal height between ion species is altered when the signal channel is changed, highlighting insufficient dynamic range of the ion detection system.

Mass-scale calibration over a limited mass range (m/z 23–56) was possible in positive ion mode due to the limited number of peaks that were universally available in all spectra, whereas a mass-scale calibration with a wider mass range (m/z 16–76) was possible in negative ion mode. Calibration coefficients obtained for PSL, Carbon black, K-feldspar and borosilicate glass are shown in Table 6. Note the increase in the intercept value for the K-feldspar and borosilicate glass samples, indicating an overall shift in the ion arrival times compared to the PSL.

The average peak position and peak width (±2 SD) of the Al+, K+ and SiO+ positive ion species and O-, SiO2- and SiO3- negative ion species in the K-feldspar, illite IMt-2 and borosilicate glass samples are displayed in Fig. 5. In positive ion mode, the peak position is often >±0.5 Da of the K-feldspar calibrated mass position in all sample types. In negative ion mode, the average position of the SiO2- and SiO3- is significantly larger for the illite IMt-2 than for K-feldspar or borosilicate glass. In addition to the peak analysis, average peak centroid and peak width are important for the creation of stick spectra used in routine cluster analysis techniques. The data-processing software provides an additional facility with which to apply a unique 3-parameter fit calibration to each spectrum. However, for this calibration to be successful, the peak of interest must be located ±0.5 Da from the average mass position, which is not the case in a significant number of peaks in mineral dust spectra.

The mass-scale dependence of the peak position shift suggests that the differences in calibrations do not arise from a simple linear shift in ion arrival times for all ion species in the mass spectra. This is further demonstrated by examining ΔT(i) with respect to the negative ion calibration for borosilicate glass (Fig. 6). Carbon black has a clear mass dependence with respect to this calibration. Kaolinite does not have a strong mass dependence but is affected by a large kink in the mass scale between the SiO2 and SiO3 molecular ions (m/z 60 and 76 respectively), which is apparent with all the silicate minerals analysed.

In this section we explore the relationship between the ion arrival times of O- elemental ion (TO) and the SiO3- molecular ion (TSiO3) within discrete ionisation events. In the case of borosilicate glass, an amorphous silicate material, a scatter plot of arrival times measured in individual mass spectra display an approximately linear distribution of data points with a gradient of 1 (Fig. 7a), indicating that ΔTO≈ΔTSiO3. In contrast, the non-silicate carbon black, a similar plot of elemental vs. molecular carbon ions measured from C at m/z -12 and C6 ion at m/z -60 gives a linear distribution of ion arrival times with a gradient of 0.35, indicating that generally ΔTC≈0.35×ΔTC6.

Similar distributions of ion arrival times can be observed with the LDI of crystalline silicate mineral dust. The framework silicate orthoclase feldspar has a distribution of ion arrival times that is similar to the distribution of the borosilicate glass but is narrower (Fig. 7b), whereas illite IMt-2, a sheet silicate, has a large mode that is similar to the carbon black, with an additional mode of particles that is more similar to the borosilicate glass. The silicate mineral sample with the most variation in ion arrival times is the illite NX for which several distinct modes are apparent (Fig. 7c). The multimodal nature of this distribution was still apparent after the particles were mass selected with the CPMA before analysis with the LAAP-TOF. The ion arrival time distribution of mass selected 350 fg illite NX particle shows a smaller variation of up to 20 ns, whereas mass-selected 700 fg illite NX particles have a distribution comparable to the polydisperse analysis.

In order to directly compare the ion arrival time distributions of different silicate minerals, the shift in arrival time (ΔT(i)) was calculated for each spectrum with respect to a set point on the timescale. For silicate-containing particles, ΔTO and ΔTSiO3 were calculated with respect to the point at which the mode distributions converge on the scatter plots at TO=4010.2 ns and TSiO3=8722.4 ns, which we will call the convergence point. The relative difference in the ion arrival times of the elemental and molecular ions in negative ion mode can then be expressed as a ratio (τ) for each particle analysed. τ=ΔTOΔTSiO3 Histograms of τ values measured from the peak analysis in silicate mineral mass spectra are shown in Fig. 8. Potassium-rich clay minerals illite NX and illite IMt-2 (Fig. 8b) have a distinct mode around 0.3–0.5, which is aligned with the ΔTC/ΔTC6 ratio derived from the peak analysis of carbon black. Other distinct modes can be seen in the illite IMt-2 sample at 0.79 and in the illite NX sample at 0.83 and 1.10, which are more similar to the distribution of τ values measured for borosilicate glass. (Fig. 8a) The smectite and montmorillonite clays have modes in the distribution that range from 0.4 to 0.86 and the 1:1 layer clay kaolinite has the largest mode at 0.93 (Fig. 8c).

The laser power setting is an important parameter in SPMS because, along with the size of the focal point and pulse duration, the amount of energy contained in each pulse defines the peak power density that occurs in the ionisation region. Differences in power density have been shown to affect the mass spectral patterns produced. For example, Reents and Schabel (2001) found that variation in peak power density, achieved by varying the 193 nm laser power setting, resulted in variations in the sodium fraction reported in the mass spectra of NaCl. The effect of different laser power settings on the distribution of τ values for kaolinite sample KGa-1b and illite NX is shown in Fig. 9. For the kaolinite-rich sample, increasing the pulse energy results in a narrowing of the distribution. The effect on the illite rich sample is somewhat different, in that increasing the pulse energy has the effect of increasing the number of particles in the mode τ>1.

An explanation for this behaviour may be found in the hit-rate efficiency, defined as the number of spectra acquired divided by the number of particles detected (Table 7). The hit rate for illite NX was more than double that of KGa-1b at 4.5 mJ. In addition, hit rate was much more sensitive to energy setting with KGa-1b than with illite NX. Increasing the laser energy from 2.8 to 7.6 mJ per pulse resulted in a hit-rate increase of 42 % for the KGa-1b sample. Species-dependent hit rates are associated with the power density threshold required to initiate the ablation/ionisation process, which is related to the lattice energy and absorbing properties of the material at 193 nm (Thomson et al., 1997). It is therefore reasonable to assume that the increase in the τ>1 fraction in illite NX at 7.6 mJ per pulse is partly due to the presence of a material with a high-power density threshold such as kaolinite. The increase in the hit rate with KGa-1b at 7.6 mJ was in a large part due to the acquisition of spectra showing a purer form of kaolinite as indicated by the titanium content (Table 7). It is not known if the titanium was present in these particles as a structural cation, therefore altering the crystal structure, or as free titanium impurity.

The main physiochemical difference between clay minerals is the nature of the interstitial X-ion that is accommodated between the silicate structural layers. For the source clays, the relative proportion of low-valence interstitial cations (X-ion) present in the structure is discernible from the structural formulae provided by the Clay Mineral Society (Table 2). These ion proportions are listed alongside the literature values for interlayer charge and value of τ of the principal mode derived from this study in Table 8. There is some negative correlation between τ and the layer charge, with the exception of the calcium-rich montmorillonite STx-1. In calcium-rich montmorillonite, the layer charge is balanced by abundant divalent ions of calcium, which have a first ionisation energy (589.8 kJmol-1) that is higher than the aluminium (577.5 kJmol-1) in the silicate structural layer. In contrast, SWy-1, ISCz-1 and IMt-2 samples have a layer charge that is principally balanced by sodium and potassium ions that have first ionisation energies that are lower than any component in the silicate structural layer (495.8 and 418.8 kJmol-1 respectively).

A comparison of the interstitial X-ion content derived from elemental analysis and the measured τ value for clay minerals and feldspars is given in Fig. 10. The combined sodium and potassium content derived from the calculated structural formulae for source clays is plotted against the mode value of τ derived from the peak analysis. For clay samples that exhibit multimodal τ distributions, only the lowest principal mode is represented in the data. Relatively high levels of sodium and potassium result in a low τ value for these modes. For feldspars, the cation content is measured from the positive ion mass spectra and plotted against τ on a single-particle basis for orthoclase and plagioclase in Fig. 10a and b. In contrast, higher sodium and potassium content in feldspars result in high τ values. The influence of the calcium ion on the τ value in plagioclase feldspar is demonstrated in the colour function in plot Fig. 10b. When Ca/Na ratios are high, the τ is >1. This has similar response to the kaolinite sample, which is depleted in interstitial X-ions.

Mass selecting 700 fg orthoclase particles with the CPMA produces a similar distribution to polydisperse particles, suggesting there is not particle size effect to the distribution. The low recorded potassium and sodium content in some feldspar particles may be a result of the crushing of the sample as this is known to cause collapse of the interstitial cavity (Garcia-guinea and Correcher, 2000) so that feldspar particles with low cation content may not be representative of naturally occurring feldspar.

Distinctive clusters of data points are apparent in the illite NX sample when τ is plotted against alkali metal content derived from single-particle mass spectra (Fig. 10c). The concentration of particles in the region τ=1.0–1.2 are clearly distinguishable from particles in the τ=0.6–1.0 region by the content of low-valence potassium and sodium ions. These clusters of data points occupy similar areas of the plot to kaolinite and feldspars. The large cluster of particles with τ<0.64 is congruent with the principal mode in the Imt-2 and ISCz-1 samples and is also characterised by relatively large peak width values. Multimodal peak shapes, as described in Sect. 2.4, mainly occur with τ > 0.64 in the illite NX sample but were never observed for feldspar with peak widths of the SiO3 molecular ion that were always < 15 ns. Broad (> 15 ns), multimodal spectral peak morphologies were attributed to aggregated or mixed particles in which the ion formation process is not uniform.

The dependence of the ion arrival times on the nature of the cations in the interstitial complex gives rise to the possibility of using these measurements to classify the mineralogical contents of mineral dust samples on a single-particle basis. Using published XRD analysis data as a guide, the classification scheme outlined in Table 9 was used to quantify the particle number concentrations of mineral types in the clay-rich mineral samples illite NX, IMt-2 and ISCz-1 (Fig. 11). The first class of particles is defined as the group of particles with τ<0.64 which only appear in samples rich in illite clays. This class is attributed to illite–smectite clay minerals (ISCM) because in natural clay samples, illite occurs in fine interlayers with smectite clays. The second class is particularly well developed in the illite NX and IMt-2 samples and is defined by τ>0.64 and high potassium and sodium content that is typical of the orthoclase (K-feldspar) sample. The exception is particles within this group that exhibit broad multimodal peak morphology which are placed in a separate class of mixed particles (class 5). Kaolinite (class 3) has, by definition, very low potassium and sodium content and is also defined by high τ measured in the pure form of the mineral (KGa-1b above). The fourth mineral class assigns particles with intermediate values of τ and relatively low potassium and sodium content to montmorillonite or Na-feldspar (plagioclase) because of the difficulty in distinguishing between these two minerals with the peak analysis. In cases in which the peak analysis was unable to produce a result, mainly due to failure of the peak fitting, were allocated to a non-classified particle class. Examples of raw mass spectra of each particle class are presented in the Supplement (S2) alongside example spectra for all other mineral samples used in this study.

The differentiation of mineral phase by this classification method has some broad agreement with the XRD analysis. The relative particle number fraction of ISCM between the three samples is in agreement with the relative mass concentration of illite derived from XRD; i.e. ISCz-1 is the most illite-rich followed by IMt-2 and illite NX. A greater fraction of feldspar and kaolinite is also measured for illite NX than for the other samples in agreement with XRD analysis. The main difference in the mineral fraction reported by XRD analysis and this peak analysis technique is the relative fraction of ISCM minerals with non-ISCM mineral classes. The differences may arise from comparing a number-counting technique with a mass fraction, and from hit-rate bias in the LAAP-TOF measurement which may discriminate against certain particle types as discussed in Sect. 3.3 above. In addition, the XRD analysis technique does not report the fraction of amorphous material and the degree of homogeneity in the bulk samples is not well characterised and may result in sample-to-sample differences in the composition of a clay mineral standard.

The relatively poor resolution of the mineral differentiation makes the identification of the exact mineral phase difficult. Whilst the plots of τ vs. K + Na content (Fig. 11, top panels) produce distinct particle clusters for illite NX and IMt-2, the corresponding plot for ISCz-1 is much less differentiated. It is not clear if this occurs because of the poor resolution of the method or because the natural variation in composition of the latter sample.

The ratio ΔTK/ΔTO is plotted against ΔTO/ΔTSiO3 (τ) in Fig. 12. In this plot, the distribution of data points for borosilicate glass and orthoclase feldspar suggests a positive correlation between these ratios, with the trends in the distributions converging towards a point at which the magnitude of shift of all species is equal. In illite IMt-2 the ISCM mineral fraction, identified as class 1 in the classification scheme, does not show a correlation between positive alkali metals ion and negative silicate molecular ion arrival times, which may indicate a decoupling of the positive and negative ion formation process.