Determination of zeolite-group mineral compositions by electron probe microanalysis

Abstract A new protocol for the quantitative determination of zeolite-group mineral compositions by electron probe microanalysis (wavelength-dispersive spectrometry) under ambient conditions, is presented. The method overcomes the most serious challenges for this mineral group, including new confidence in the fundamentally important Si-Al ratio. Development tests were undertaken on a set of natural zeolite candidate reference samples, representing the compositional extremes of Na, K, Cs, Mg, Ca, Sr and Ba zeolites, to demonstrate and assess the extent of beam interaction effects on each oxide component for each mineral. These tests highlight the variability and impact of component mobility due to beam interaction, and show that it can be minimized with recommended operating conditions of 15 kV, 2 nA, a defocused, 20 μm spot size, and element prioritizing with the spectrometer configuration. The protocol represents a pragmatic solution that works, but provides scope for additional optimization where required. Vital to the determination of high-quality results is the attention to careful preparations and the employment of strict criteria for data reduction and quality control, including the monitoring and removal of non-zeolitic contaminants from the data (mainly Fe and clay phases). Essential quality criteria include the zeolite-specific parameters of R value (Si/(Si + Al + Fe3+), the 'E%' charge-balance calculation, and the weight percent of non-hydrous total oxides. When these criteria are applied in conjunction with the recommended analytical operating conditions, excellent inter-batch reproducibility is demonstrated. Application of the method to zeolites with complex solid-solution compositions is effective, enabling more precise geochemical discrimination for occurrence-composition studies. Phase validation for the reference set was conducted satisfactorily with the use of X-ray diffraction and laser-ablation inductively-coupled plasma mass spectroscopy.


Introduction
ZEOLITE-group minerals are hydrated aluminium silicates of the alkali and alkaline earth elements, with open framework structures of linked (Si,Al)O 4 tetrahedra (Passaglia and Sheppard, 2001;Deeret al., 2004). They are well known for their properties as ion exchangers and as 'molecular sieves' (Dyer, 1988), and natural zeolites are exploited widely as industrial minerals in applications as diverse as soil amendments, swimming pool filtration, wine making, pharmaceuticals (Dakovićet al., 2014), and nuclear waste management (Dyer, 2000). In addition to natural varieties, hundreds of synthetic zeolites and similarly-structured materials of other compositions ('zeotypes') have become well established for highspecificity applications in petroleum and material sciences (Čejka et al., 2010).
The scientific value of zeolites in the Earth sciences is found partly in their sensitivity to different temperatures and geochemical environments (Chipera and Apps, 2001), enabling predictive deductions to be made about formation conditions from mineral assemblages and compositions (e.g. Neuhoff et al., 2000;Bish and Ming, 2001;Giampaolo et al., 2008;Campbell et al., 2012;Langella et al. 2013;Weisenberger et al., 2014;Cappelletti et al., 2015). This has particular relevance to hydrothermal, surface and diagenetic systems, where mineral compositions can track changing fluid regimes (Hay and Sheppard, 2001;Langella et al., 2001;Utada, 2001, Campbell et al., 2012Langella et al. 2013;Weisenberger et al. 2014). Where natural mineral reaction paths involving zeolites are combined with other geological criteria, they have potential value in mineral exploration (Campbell et al., 2012(Campbell et al., , 2013(Campbell et al., , 2014, in deepening understanding of global carbon cycling and climate change (Heister et al., 2001;Campbell et al., 2012) and in contributing to the rationalization of volcanic hazard processes (Giampaolo et al., 2008;Bear et al., 2009;Vignaroli et al., 2014). Further, mineral reactions involving zeolites are relevant to studies on the fate of nuclear waste (Wallace et al., 2013). In all these cases, a robust method of mineral analysis is required to achieve full, high quality and reproducible data sets of zeolite mineral compositions, for use in geological and geochemical interpretation based on better understandings of atomic substitutions (Neuhoff and Ruhl, 2006;Gatta et al., 2009). The primary objective of this study therefore, was to generate and validate a robust protocol for the routine determination of zeolite-group mineral compositions by electron probe microanalysis (EPMA, or 'microprobe'). A secondary objective was to select and characterize a set of reference zeolites representative of the compositional extremes of this mineral group, for future quality assurance in analysis.
It is well known that there are severe and inherent problems with the analysis of zeolite-group minerals using micro-beam methods of X-ray generation (Henderson et al., 2014;Weisenberger et al., 2014). Quantitative detection of the resulting characteristic X-rays is either by energy-dispersive spectrometry (EDS), or by wavelength-dispersive spectrometry (WDS). The main problems arise from the interaction of the electron beam with the sample, causing heating (thermal properties of zeolites are reviewed in Bish and Carey, 2001), and diffusive mobility of compositional components in and around the analytical volume, as demonstrated in Kearns and Buse (2012), for alkalis in volcanic glass, and further potentially explained by charge implantation (Cazaux, 2004;Fakhfakh et al., 2010). The mobility includes dehydration and various effects on the light elements, Na, Mg, Al, Si, as indicated in related studies (Line et al., 1995;London, 1996, 2005;Vaggelli et al., 1999;Deer et al., 2004;Putnis et al., 2007;Rigby et al., 2008;Weisenberger and Spürgin, 2009;Kearns and Buse, 2012;Henderson et al., 2014;Pearce et al., 2014). These studies, though mostly on glasses and non-zeolitic mineral phases, bear relevance to the EPMA analysis of zeolites. This is because the extra-framework cations in zeolites, especially the lighter alkalis, are very loosely bound in exchangeable sites, with complex crystal-chemical relationships with lattice H 2 O and with the Si-Al oxide framework (Passaglia and Sheppard, 2001). Atoms in natural glasses have some level of structural co-ordination (Si-Al tetrahedra, as polymerized 'networks' with interstitial cations which may, or may not, be hydrated), regardless of having no crystal structure, as such. A critical issue for zeolite-group minerals is in the loss of Na with concomitant differential 'grow-in' of Al and Si (as described by London, 1996, 2005 for glasses), and the limitation that this places on determination of the Si/Al ratio, which is of fundamental importance in the quantification of zeolite compositions (Passaglia and Sheppard, 2001;Neuhoff and Ruhl, 2006). The general understanding of Al + Si 'grow in' is that as the light and volatile components are driven away from the heated interaction volume (in our case, of the zeolite structure), the apparent, relative concentrations of Si and Al increase. Morgan and London (1996) also explain that the outward migration of Na causes a reduction in the absorption of emitted X-rays from the other elements, and that this effect is greater for AlKα than for SiKα due to relative differences in mass absorption coefficients.
Further, it is possible that dynamic interactions between H 2 O and extra-framework cations influence framework bond strengths and angles during dehydration, potentially leading to phase transitions Bish, 2012, 2014). Electron beam interaction can also affect the apparent K signal in specific zeolite minerals, notably chabazite, as we report here. Time-dependent intensity (TDI) calculations are referred to by Morgan and London (2005) as a possible way to deal with alkali migration issues in EPMA. However, it is considered that errors introduced through the assumptions and estimates that are needed for this method would probably be much higher than the negligible, measurable errors resulting from element prioritizing in the analysis protocol. The results shown in Test 4 confirm that element prioritizing is a reasonable method, and that TDI is not required. With the advent of field-emission gun (FEG) technology in new-generation microprobes (Merlet andLlovet, 2012, Saunders et al., 2014), it is anticipated that very significant future improvements relating to beam interaction problems will be possible for zeolite analysis, allowing much higher spatial resolution studies to be undertaken.
An additional problem for EDS methods is in the peak overlaps of Si-K and Sr-L X-ray emission peaks, again precluding the effective determination of Si/Al due to the presence of Sr, which can be an important element in many natural zeolites (Passaglia, 1970;Pekov et al., 2000;Deer et al., 2004;Campbell et al., 2015), and which also requires measurement. In this present study, we address all these problems with tests using a WDS method (non-FEG EPMA, with a 5-spectrometer Cameca SX100 microprobe), on a representative set of candidate reference zeolites. Based on our findings, we present recommended, pragmatic protocols for analysis and data processing for the robust, yet routine, ambient temperature quantification of zeolite compositions by conventional EPMA.

Operating conditions
Initial decisions on the microprobe instrumentation set-up and operating conditions were based on the published literature (Morgan and London, 1996;Giampaolo et al., 1997;Vaggelli et al., 1999;Morgan and London, 2005;Rigby et al. 2008;Weisenberger and Spürgin, 2009;Kearns and Buse, 2012;Weisenberger et al., 2014;Henderson et al., 2014;and Danisi et al., 2015). The cryometric method of Kearns and Buse (2012) is acknowledged as a highly effective way of minimizing the heating effect of beam interaction, but it is a specialized technique that is not widely accessible in many EPMA laboratories. Instead, therefore, ambient temperatures were used in the current study, optimizing the other important parameters of beam current, spot size, and order of detection of elements in the spectrometer set-up (see Table 1 for main analyses and Table 2 for X-ray mapping), with the most susceptible elements Na, Al, K, Cs and Ca being analysed first and simultaneously on five separate spectrometers to minimize problems. The beam conditions were first set to 15 kV, 4 nA and 20 μm (defocused), and later set to 2 nA for our established protocol. The selection of an accelerating voltage of 15 kV represents a compromise between limiting the beam penetration depth and corresponding light-element X-ray absorption effects (which increase with increasing kVsee Merlet and Llovet, 2012), and limiting the thermal effect of a higher electron density with a smaller (shallower) volume of interaction at lower kV (Morgan and London, 2005). Nevertheless, the complexity of beam interaction effects on alkali silicate materials as indicated by Cazaux (2004), suggests that there is future scope for examining the effect of varying the kV specifically for zeolites. This is supported by a useful review explaining a number of parameters that are affected by low kV EPMA, but mainly relevant to FEG-EPMA (Merlet and Llovet, 2012).
Tests using each of the other operational configurations in the present study were run on a set of natural zeolites chosen to represent all key compositional 'end-members' (See the section on Reference samples and Table 3, described below), and the data were then evaluated for understanding optimal set-up parameters. One of the most important types of test was count-rate monitoring at intervals of 200 ms over 300 s, per element, per sample, as this would reveal the differential extent of sensitivities to beam interaction (see Test 4).

Calibration
Instrument calibration standards for the Manchester Cameca SX100 microprobe relevant to this study are listed in Table 1. These standards, and many others, are employed on a regular basis for routine weekly instrument calibration and used by all analysts. A wider set than listed in Table 1 was further used in the present study with our run file to assess for potential unresolved interferences, and therefore they also serve as 'interference blanks'. The findings are described in Test 1.

Element set-up
The five-spectrometer configuration for our analytical protocol is given in Table 1. Based on the recommendations of London (1996, 2005), and on recent studies of natural zeolites (Langella et al., 2013;Weisenberger et al., 2014;Henderson et al., 2014), the priority was to analyse Na and Al first (TAP detectors) with K and Ca (PET detectors), and Cs (LIF) simultaneously, followed immediately by the other elements (Si, Mg, Sr, Ba and Fe). Rubidium was monitored in initial runs, but never detected (limit of detection generally <0.06 wt.% Rb 2 O for most zeolites, and 0.21 wt.% Rb 2 O for pollucite). Apart from in leucite and clinoptilolite, where minor Fe 3+ is known in  tetrahedral co-ordination (Gatta et al., 2007, for leucite data), Fe in most zeolites is considered to be a contaminant (Passaglia and Sheppard, 2001). It is therefore included in the element list in our analytical protocol as Fe 2 O 3 for monitoring of contamination, hence forming one of the decision-criteria for exclusion of low-quality data in our Appendix: Data Reduction and Quality Control Protocol. These checks are necessary in variably-hydrated minerals in which a range of low total oxides can be expected, and they help to minimize uncertainty about the elements that can not be analysed. However, while all analyses in this study are reported on a volatilefree basis with H 2 O determined by difference (it can reach to over 20 wt.% in zeolites, Deer et al., 2004), the possibility of minor carbonate or bicarbonate on internal and external mineral surfaces remains (Harjula et al., 1993;Dyer, 2007). Dyer (2007) and Dyer et al. (2010) speculated that the unusual ion exchange behaviour of high selectivities of Sr in high-silica zeolites under alkaline conditions could be due to monovalent speciation [SrX] + , (where Xmay be OHor HCO 3 -). The selectivity is unusual, based on fundamental principles of compositiondependent silicate-solute interactions as discussed in Eisenman (1962) and explained in Colella, (1996), and in Dyer (2000) for zeolites. Hence, low-chargedensity (high-Si) zeolites are predicted to show a preference for large, monovalent cations, supporting the notion of the potential presence of anions other than O 2-. Carbonate sorption on synthetic zeolite surfaces is discussed in Harjula et al. (1993), who demonstrate how the reactive interplay of NaOH -, HCO 3 and CO 3 2solutes affects the Si-Al framework. For low-Si zeolites, there has been recent interest in the chabazite structure and its affinity for CO 2 , in relation to carbon-capture studies using synthetics (Kim et al., 2014;Pham et al., 2014). Thus, while it is difficult to analyse for low concentrations of C routinely by EPMA, its potential presence should be borne in mind in interpretation of compositional data, and its determination by other analytical techniques should be considered (e.g. vibrational spectroscopies for carbonate components). Beryllium, B and Li are also known in some zeolites (Deer et al., 2004), the first two occupying framework tetrahedral sites. However, these rare compositions are not considered in the present study.

Reproducibility
When the optimal operating conditions for zeolitegroup minerals were determined, selected reference samples were analysed repeatedly in separate sessions (different days), to monitor inter-batch reproducibility. This was especially of value for the solid-solution compositions (levyne, phillipsite and chabazite), as shown in the results for Test 5. All reference samples were evaluated for homogeneity with the combined use of replicated spot EPMA analyses and of X-ray maps (described in the 'Reference samples' section, below).

Data processing
The raw WDS data were scanned for poor quality analyses on the basis of the following criteria, applied iteratively: (1) Presence of contaminants (Fe 2 O3 > 0.2 wt.%, and in initial studies, P 2 O 5 > Limit of Detection, 'LOD'); (2) application of the 'E%' error test of Passaglia et al. (1970), omitting analyses with E% less than -10 or greater than +10; (3) totals of non-volatile oxides outside the range 80-95 wt.%, except for leucite (anhydrous, ideally 100%); and (4) R value, Si/(Σ TET ) (tetrahedral framework cations), outside the ranges listed in table 1 of Passaglia and Sheppard (2001). Once all these criteria were applied, acceptable analyses were then processed for limits of detection, per oxide, per mineral, per session. Deletion of poor data then continued after recalculation of total oxides and E%. Calculation of mineral formulae was undertaken according to the conventions set out in Deer et al. (2004) based on numbers of framework oxygen atoms specific to each different mineral species. For example, 24 oxygens for chabazite, 20 for edingtonite, 32 for phillipsite, brewsterite and goosecreekite, etc. The H 2 O content (wt.%) was determined by difference.
Stepwise recommendations using our protocols are given in Appendix 1.

Validation by XRD
Where possible, sub-samples of the reference zeolites used in the EPMA study were independently characterized by X-ray diffraction (XRD). This was primarily for confirmation of mineral phase identification, and the following method was used: Specimens of a few tens of milligrams were hand ground in an agate mortar and pestle for 2 min and the resulting powders sprinkled onto low background silicon single-crystal foil. Specimens were scanned using Ni-filter Cu radiation from 3-70°2θ, in 0.0167°steps counting for 100 s per step using a Panalytical X-pert Pro instrument fitted with an X-celerator position-sensitive detector. Mineral identification was performed using Bruker Diffrac EVA software and comparison to patterns in the Powder Diffraction Files of the International Centre for Diffraction Data: (ICDD PDF-4 2013 database).

Validation by LA-ICP-MS
For compositional validation, there was an opportunity to use major-element data that had been part of a separate, preliminary investigation for traceelement capability using laser ablation inductivelycoupled plasma mass spectroscopy (LA-ICP-MS). Samples were analysed using an Agilent 7500 ICP-MS system at the British Geological Survey (Keyworth, UK). This is equipped with a New Wave UP-193 laser ablation system fitted with a 193 nm excimer laser (New Wave Research, Fremont, California, USA), and controlled by New Wave Research-Laser Ablation software. The ablated material is transported in a continuous flow of He to the ICP-MS, where it is vapourized and ionized at 6000 K. Determinations of multi-element mass concentrations were based on calibrations using standard glass NIST612 and an internal 44 Ca standard. The method is conveniently capable of near-simultaneous measurement of both major and trace elements, but has limitations in trace-element sensitivities and in analysis volume. Detection limits are proportional to crater volume; to ensure signals above detection limits for some lowabundance elements, craters as large as 100 μm are sometimes required, as in the case of our preliminary trials with zeolites. Another issue is that there are potential interferences of 44 Ca + by 12 C-16 O-16 O + derived from carbonate or mounting resin, of Ba on the middle rare-earth elements ( 137 Ba 16 O + / 153 Eu + ), and mutual doubly charged and polyatomic interferences between Ca and Sr ( 44 Ca + / 88 Sr 2+ and 44 Ca-44 Ca/ 88 Sr 2+ ), limiting potential applications of this particular set-up. All analyses were normalized to the mean SiO 2 wt.% determined by EPMA.

Reference sample selection
One of the key purposes of this study was to compile and characterize a set of reference zeolites for long term use in analytical quality control, as no suitable certified reference materials have been available previously. Whilst there is an abundance of synthetic zeolites available for industrial uses, they are all microcrystalline and hence, unsuitable for the present purpose (Fischer, 2014). The availability of single crystals from natural samples therefore remains as providing the best opportunity for developing sets of reference data. The criteria for selection of our candidate reference minerals of trusted provenance were: (1) representation of compositional end-members by extra-framework Na, K, Cs, Mg, Ca, Sr and Ba; (2) representation of different zeolite structural types; and (3) ideally, a minimum crystal size of ∼1 mm 3 . All the common extra-framework cations are represented in our candidate reference samples (Table 3 and Supplementary file 1 that has been deposited with the Principal Editor of Mineralogical Magazine and is available from www.minersoc.org/pages/e_ journals/dep_mat_mm.html). The pollucite for Cs (Cs 2 Al 2 Si 4 O 12 .H 2 O), is of unknown provenance, but was available in the set of microprobe calibration standards at Manchester University. Here, we simply report its analysed composition based on our new protocols. Whilst leucite for K is now formally classified with the zeolites, it has limited value as a reference material for zeolitic K because it is not hydrated (see Henderson et al., 1998, for discussion of the crystal chemistry of the leucite-pollucite-wairakite structure type). Henderson et al. (2014) showed that the NaAlSi 2 O 6 .H 2 O component of leucites can be up to a maximum of 4 wt.%. Nevertheless, the reasonably consistent formula and lower Si:Al of leucite than of K-feldspar renders it useful in conjunction with the solid solutions levyne (sbW), chabazite (sbZ) and phillipsite (Z20). The faujasite (fbH for Mg-reference) and brewsterite (Z14 and sbP for Sr-reference), similarly, are solid solutions but necessarily part of the reference set due to the absence of pure Mg and Sr end-member zeolites in nature. The inclusion of natural zeolite minerals which display solid-solution ranges was considered important as these minerals are reported widely in occurrence studies and have considerable value in geochemical discrimination when used in conjunction with paragenesis information (e.g. Heister et al., 2001;Weisenberger and Spürgin, 2009;Weisenberger et al., 2012Weisenberger et al., , 2014Langella et al., 2013;Campbell et al., 2015;Cappelletti et al., 2015). The solid-solution samples additionally have value in monitoring reproducibility (see Test 5). A future objective is to acquire and characterize a suitable heulandite or clinoptilolite reference sample, representing high-Si zeolites that display complex solid-solution systematics. Clinoptilolite deposits occur abundantly and are economically important as industrial minerals. Examples of geologically-relevant structural-type differences in the reference set are provided by natrolite and analcime for Na end-members, and wairakite, laumontite and goosecreekite for Ca end-members.

Reference sample characterization
Details of the selected candidate reference samples and their sources are given in Table 3 and Supplementary file 1 (Supplementary files 1-4 are deposited at www.minersoc.org/pages/e_journals/dept_mat_mm.html), and diagrams of their mounted positions in the polished sections are provided in Supplementary file 2, for potential long-term reference. Optical and electron microscopic images were collected for navigation and quality evaluation purposes, and X-ray maps were collected to assess compositional homogeneity. All of these images are compiled in Supplementary file 3. The phillipsite (Z20) and one of the brewsterites (Z14) did not display adequate compositional homogeneity for full use as secondary standards, but in the absence of any other Sr-rich zeolite, the brewsterite nevertheless has value as a basic-level reference material, as discussed above, and in the analysis development tests of the present study (below).
Averaged, optimum analyses for each mineral determined using our final, recommended protocol, are shown in Table 4, but the full set of EPMA analyses based on a variety of test parameters are given in Supplementary file 4, with limits of detection. The detail and pattern of results for each development test are explained below. Where available, analyses are compared with those from previously published data, as detailed in Deer et al. (2004) and in other more recent publications. Note the excellent 'error test' (E%, Passaglia, 1970) results and low variances of the new data in the summary table (Table 4). Passaglia (1970) recognized the scope of a simple arithmetic test for zeolitic analytical quality, by using the mineralogical constraint of the charge equivalence of Al 3+ (and the associated anionic excess in the [Si,Al]O 4 framework), with the sum of the charges of the extra-framework cations. Thus, an E% value of zero represents full stoichiometry (excluding H 2 O). The general guidance for acceptable analyses is that they should fall within the E% range -10 to +10%, but even better, within the range -7 to +7%. In Table 4, most data fall within -2 to +2%, with only natrolite Z01 outside this very high quality range at -5.32% and a 'replacement', or 'X-type' analcime (Giampaolo and Lombardi, 1994;Henderson et al., 2014) at +7.5%. Error test results in Supplementary file 4 should be read in conjunction with the appropriate descriptions of the development tests, explained fully, below.
Reference sample validation XRD data Mineral phase identifications are shown in Fig. 1, with spectral traces in colour superimposed onto the ideal (ICDD) peak positions and intensities in grey/black. The results serve as verification for all phases determined this way.

LA-ICP-MS data
In Table 5, averaged LA-ICP-MS results for the major elements are presented, with normalization to the mean wt.% SiO 2 as determined by EPMA. There is excellent agreement of results for three candidate reference materials natrolite (fbN), brewsterite (sbP) and levyne (sbW), except for Ca (extrapolated from 42 Ca) in the LA-ICP-MS data, relative to the microprobe data (Table 4 and Supplementary file 4).

Development test results and discussion
Test 1: X-ray interferences Following an evaluation of potential peak overlaps using 'Virtual WDS' software, and adjusting background positions and spectrometer choices accordingly, this first test examined the quality of 'blank' runs acquired on the calibration standards. All blank runs provided satisfactorily low blanks for zeolite-relevant major elements. The apparent 0.5 ± 0.15 wt.% Al 2 O 3 seen for the barite standard is unexplained but assumed to be a contamination issue rather than an interference, as no similar signal was found for the benitoite (BaTiSi 3 O 9 ), the anhydrite or the celestine standards (relating to Ba and S components respectively). The potential Ti-Ba X-ray interference was tested with the rutile standard satisfactorily. In our protocol, Ba-glass was used as the calibration standard for Ba (Table 1). Although Cs-glass was used as the calibration standard for Cs, a pollucite standard of unknown source was also available, and it was decided that this could serve as the Cs end-member reference mineral for the present study. Analyses of the pollucite standard were therefore collected, and presented with all the other results in Table 4 and Supplementary file 4.
Test 2: Effects of beam interaction on basic reproducibility for different minerals.
In this test, multiple analyses were undertaken for a single spot, using a defocused beam of 20 µm, on different minerals. Operating conditions were 4 nA and 15 kV, and two test minerals were used; natrolite Z01 and mesolite Z03. For comparison, similar data for the albite and jadeite standards  (non-hydrated silicates) were obtained. Figure 2 shows the change in apparent oxide wt.% for successive analyses on the same spot. The greatest changes with time are seen for Na 2 O and Al 2 O 3 in natrolite after the fourth analysis, but steady and immediate losses of Na 2 O are seen for mesolite. SiO 2 is generally stable except in natrolite where a small but perceptible change is apparent in the later analyses. These initial observations are consistent with many previous reports of Na loss and Al + Si 'grow in', in hydrous, alkali aluminosilicate glasses, as reported by London (1996, 2005). Potential explanations, including heat-driven diffusion (Kearns and Buse, 2012), have been outlined in our introductory paragraphs. Therefore, in consideration of the evidence outlined above for differential effects of beam interaction specific to different elements and different phases, further element-specific and mineral-specific tests were conducted, as follows.

Test 3: Effect of beam current
Adhering to a beam current of <5 nA, based on the findings of Morgan and London (1996), the third set of tests compare analyses determined at 4 nA with those determined at 2 nA, for ten separate reference minerals: leucite, analcime (2), natrolite (3), mesolite, edingtonite (2) and brewsterite; Supplementary file 4. In Fig. 3a-c, it can be seen that the 2 nA populations for leucite, X-type analcime (Z06) and mesolite tend to perform better in terms of stoichiometry (based on the E% and on total wt.% oxides), than the 4 nA populations, but that for H-type analcime (Z04), there was no difference with current (2 vs. 4 nA). For analcimes, a greater influence on the quality of the analysis is in the crystallization history and apparent susceptibility to factors controlling permeability within the crystal. The H-type analcime (Z04) from Dean Quarry (Lizard Complex, Cornwall, UK) is from a pegmatitic/hydrothermal vein, presumably crystallizing directly from a fluid. The analyses display excellent stoichiometry, with H 2 O in the range 7-12 wt.%. In contrast, the X-type analcime (Z06) from Mt. Vulture (southern Italy), occurs enclosing leucite (Z05) in an altered pyroclastic deposit, and it crystallized by a replacement mechanism expressed by the wellknown ion exchange reaction accompanied by hydration:

Leucite Analcime
These X-type analcimes are commonly reported as being depleted in apparent Na content, and additionally display lower R values and extreme E% values than for an ideal analcime (Deer et al., 2004). X-type analcimes as pseudomorphs after leucite, tend to be very fine-grained and polycrystalline, giving a high surface area promoting mobility of volatiles. They contrast the single crystal, vitreous appearance of H-type (hydrothermal) and I-type primary (igneous/magmatic) analcimes Lombardi, 1994, Putnis et al., 2007;Henderson et al., 2014). Several reasons for Na-depletion in X-type, beam-affected (dehydrated) analcimes have been discussed previously, and include crystal defects, framework volume increase with fracture development, and H 2 O-Na mobility. Additionally, we suggest that the reactive expulsion of trace elements from leucite, possibly remaining in situ as micro/nano secondary-phase inclusions in the analcime product, causes further enhancement of the overall permeability, for rapid loss of volatiles on heating (beam interaction). The interface zone in the X-ray maps of the leucite-analcime pair (Supplementary file 3), showing marginally elevated Mg (probably occurring as smectite) is consistent with this, and with the replacement mechanism of dissolution-reprecipitation proposed by Putnis et al. (2007). Neither K nor Ca were detected in Z06 (Supplementary file 4, confirmed with X-ray maps of the leucite-analcime pair, Z05 and Z06, Supplementary file 3), indicating a complete absence of potential residual K that might have been derived from the primary leucite. (The apparent trace-Ca in the Z06 map is likely to be a background-matrix artefact, and the traces of leucite in the XRD data for Z06 are due to the paired co-occurrence of the two separate phases and practicalities of complete separation prior to powdering for analysis). Natrolite and edingtonite displayed no differences in analytical quality due to 2 nA vs. 4 nA beam conditions (Fig. 3d,e).
For Si:Al in all phases studied, the effect of beam current (2 and 4 nA) appears to be minimal or absent, except for where systematic coupled substitutions involving alkalis as seen in brewsterite (Fig. 3f ), suggest a potential influence. However, only two points of data collected with the 4 nA condition were accepted with our processing protocol, and these happened to display full occupancy of the EC sites by Sr 2+ and Ba 2+ alone. Nevertheless, the positive correlation of total oxides with the sum of Sr and Ba (and therefore, of alkali content and H 2 O) in the 2 nA brewsterite data, is suggestive of a preferred crystallographic association of H 2 O with the alkalis over M 2+ , potentially leading to dehydration-coupled alkali losses if higher currents were used.

Test 4: Count-rate stability
The extent and analytical impact of beam-interaction susceptibility in different zeolite minerals was investigated by count-rate monitoring tests, per sample, per element, against time at a fixed current of 2 nA. The spectrometer-element configurations were identical to those used for the main analysis run-file (Table 1). The sampling rate was every 0.2 s over a period of 300 s. Spectrometers were run simultaneously where possible, but fresh spot positions were selected when it was necessary to change spectrometer configurations for different elements in these tests. Figure 4 shows selected count rate profiles for reference sample minerals, by element. Signal stability is indicated by a nearhorizontal profile, as demonstrated by the Dean Quarry natrolite (Z01) between 0 and 1800 s. In addition to the second-by-second random fluctuations in count rates, most profiles show minor, longer-period fluctuations over several seconds or tens of seconds, but these are generally of lower magnitude than the second-by-second fluctuations, and are probably related to subtle instrument sensitivities to the immediate beam-environment.
Deviations from horizontal profiles that are not accounted for in the above rationale are examined further. Analcime Z06 (X-type) from Mt. Vulture, Italy, displays a marked decrease in count rate for Na (∼50%), and a minor increase for Si over the course of 300 s. This pattern, typical of previous descriptions of Na-loss, is not seen in analcime Z04 (H-type) from Dean Quarry, Cornwall, UK. We again account for the differences for this phase by consideration of the beam interaction issues as discussed for Tests 2 and 3, but additionally, the XRD data of Fig. 1 demonstrate crystal structure differences; X-type Z06 is cubic Ia3d or orthorhombic Ibca, whereas H-type Z04 is possibly monoclinic, based on a cubic Ia3d pseudocell. Analcime is known to crystallize in as many as 5 different crystal systems (Deer et al., 2004). The higher surface area of analcime Z06 also potentially increases its susceptibility to CO 2 sorption from the atmosphere, affecting the overall reactive potential of this Na-rich mineral in alkaline conditions, including framework dissolution (Harjula et al., 1993). Putnis et al. (2007) further show that X-type analcimes derived from leucite might undergo replacement processes that are more complex than simple ion exchange. Their evidence is based on experiments using 18 O water, indicating the involvement of framework-O exchange in the reaction. Other reference samples in the present study displaying cumulative loss of light elements include mesolite (Z03) losing Na and Ca (Test 2 only, but resolved with the new protocolssee Fig. 4a), and phillipsite (Z20) losing K (Fig. 4c). For phillipsite, only minimal 'grow-in' effects were evident in the count rate profiles for Si and Al, and then only after ∼2 min. Phillipsite (Z20) was the only test sample to display a decrease in count rate for K, but a new phenomenon of an increase in count rate for K was observed in chabazite (sbZ) (Fig. 4b). Indeed, the observation was repeated in a number of other chabazites (outside the present study), and their profiles have also been included in Fig. 4b (chabazite sbT from a Scottish vug in Tertiary basalt and chabazite IK23 from an Italian zeolitized tuff ). Faujasite (fbH) additionally displayed a count rate increase for K of ∼400% within the first two minutes of beam interaction, and this was accompanied by a count rate increase of ∼150% for Na relative to its count rate at time = 0 (Fig. 4c). A perceptible increase in the count rate for Na in levyne (sbW) was also observed. Effects in the framework Al and Si count rates for all these phases are absent or extremely marginal, and the M 2+ cations are completely unaffected under conditions for Test 4 (2 nA, 15 kV, 20 μm spot). Investigating the causes of the new observations, especially for K in chabazite, is beyond the scope of the present study but high mobilities could be associated with the large structural framework cages of chabazite and faujasite (Deer et al., 2004). The crystal chemistry of phillipsite in relation to K sites and temperature is discussed in Gatta et al. (2009).

Test 5: Solid-solution representation and reproducibility
The principal aim in the selection of most of the reference samples was to have representation of the key 'pure' stoichiometric end-members (i.e. analcime and natrolite for Na, leucite for K, pollucite for Cs, wairakite, laumontite and goosecreekite for Ca, and edingtonite for Ba), and this has been achieved. However, as it is anticipated that many of the most useful zeolites for understanding Earth process will be found in solid-solution compositions (commonly in phillipsite, chabazite and clinoptilolite), it was considered important to develop a few of these as reference materials too. For solid solutions, compositional trends for crystal-chemical inferences are best represented FIG. 4. Continued graphically, so in addition to the tables of EPMA analyses (Table 4 and Supplementary file 4), ternary plots of the extra-framework cation compositions from successive batches of data have been constructed (Fig. 5). For both the levyne (sbW) and the chabazite (sbZ) in Fig. 5a, the data points are tightly clustered within a small range, and display no apparent differences across batches. Phillipsite (Z20) data indicate a reasonably tight compositional range across analytical-session batches, but not as tightly constrained as for levyne (sbW) and chabazite (sbZ). The X-ray maps for phillipsite (Z20) demonstrate compositional heterogeneity (chemical zoning), which partly accounts for the spread of data. Using the new data, it is observed that in levyne (sbW) and chabazite (sbZ), trends of increasing M 2+ occur with constant Na + /K + ratios, but that in phillipsite (Z20), the trend of increasing K + occurs with a constant Na + /M 2+ ratio. These observations are consistent with the crystalchemical site preferences as detailed in Deer et al. (2004), and demonstrate new scope for further understanding the compositional systematics of zeolite-group minerals.
For Sr, where no end-member composition is known in nature, brewsterite from the type locality Strontian, Scotland, has provided a reasonable solid solution alternative (Fig. 5b). In brewsterite, the main substitutions are between Sr and Ba, but minor Na, K and Ca can be present, as seen in Supplementary file 4. Edingtonite is represented together with brewsterite in Fig. 5b. For Mg, faujasite from the type locality Limberg, Germany, is another complex solid-solution phase and the reference sample fbH has been characterized for similar purposes (Table 4 and Supplementary file 4). To date, only one batch of data is available for the faujasite, and the high SiO 2 content relative to reported analyses in Deer et al. (2004) might be indicative of minor clay contamination. Nevertheless, the data are otherwise reasonably consistent with those of Rinaldi et al. (1975).
Overall, these results provide confidence in interbatch reproducibility of data where complex solidsolution compositions are anticipated.

Summary and conclusions
The reference samples Similarly, both the Honshu wairakite (Z17) and the Dumbartonshire edingtonite (Z12 and Z13) display high-purity compositions, except for trace Na 2 O in wairakite (<0.02 wt.%) and minor K 2 O in edingtonite (<0.5 wt.%). The composition of the natural Italian leucite (Z05) compares well with previous studies but a variety of elements at trace levels are reported and a small analcime component is evident. Solid-solution reference materials including brewsterite for Sr, levyne for Na-Ca, and widely occurring phillipsite and chabazite have demonstrated good inter-batch analytical reproducibility, reflecting high precision for confidence in analytical quality.

Methodology and protocols
Volatile losses have been minimized with the use of carefully selected operating conditions, but it is nevertheless recommended that responses to beam interaction are monitored with each new batch and each different mineral, by means of count-rate tests, per element, on the specified WD spectrometers. This will ensure that unexpected deviations from expected count-rate fluctuations in new samples are tracked and mitigated using the monitoring data. Monitoring is also recommended because new questions have arisen from this development work, most notably with the response of K to beam interaction in chabazite and in other specific zeolite minerals. The data processing protocol is also vitally important to ensure that only high quality analyses are accepted. Overall, the key limitations are (1) that zeolite mineral compositions involving significant Li, Be and B components cannot be adequately characterized by conventional (non-FEG) EPMA, and (2) that spatial resolution remains a challenge for many very fine-grained samples such as in zeolitized tuffs. Although there is an analytical argument that some resilient zeolites rich in M 2+ cations might respond well with smaller spot sizes (e.g. brewsterite, laumontite), in practice, it is the common solid-solution zeolites (e.g. phillipsite, chabazite, clinoptilolite, faujasite, stilbite) that display alkali-bearing compositional variations, that are the most likely minerals to be of compositional-research interest, and that therefore also require the analytical conditions as recommended. The limitations relating to spatial resolution are not insurmountable and can be partly overcome with skilful petrographic preparations, but it is also anticipated that the advent of a new generation of microprobes (FEG technology) will allow the greatest step-change in analytical capability for zeolites. It has been demonstrated that with careful preparations and analysis set-up conditions, the quantitative determination of major and minor elements in most of the common natural Al-Si zeolites is achievable routinely by EPMAwithout the need for specialist cryometric instrumentation. Robust quality-assurance procedures that include reference sample monitoring and strict data reduction criteria, are recommended as vital to the achievement of high quality analyses. Due to the complexity of solid solutions and compositional ranges in zeolitegroup minerals, it is also recommended that mineral phase identification by XRD accompanies compositional studies. Stepwise recommendations for the quantitative determination of zeolite-group mineral compositions by EPMA are provided in Appendix 1.

Wider implications
With new confidence in overcoming the most challenging problems that have existed with the microbeam analysis of zeolite-group minerals, (namely, the Si-Al ratio, quantification of Na and Sr, and rationalization of the Fe issue), our protocols open new vistas for research capacity in zeolite mineralogy and Earth system sciences. Determination of zeolite-group mineral compositions will impact on understandings of volcanichydrothermal-diagenetic processes, alkaline fluid evolution with estimation of pressure, temperature and pH conditions, and element cycling, especially where mineral reaction paths are considered. These types of studies find application in natural resources (geochemical discrimination for mineral exploration; characterization of industrial zeolite deposits; nutrient status of volcanic soils), in natural hazards (volcanic-volatile processes that can impact on climate change), and potentially, in Earth analogue studies of Martian mineralogy.
1.5. Conduct count-rate monitoring tests on the phase(s) to be analysed, per element, per spectrometer, according to the spectrometer configuration in the analysis declaration (for this study, see Table 1 and Fig. 4). This is only necessary once per sample. 1.6. Analyse reference zeolites appropriate to the expected sample compositions with each new session. A minimum of five replicate analyses is recommended for pure 'end-member' compositions (e.g. natrolite, edingtonite), and significantly more for solid-solution references (e.g. phillipsite, brewsterite). 1.7. Sample points: For samples where spatial resolution and/or polish quality is not an issue (hydrothermal, cavity types), it is expected that at least 90% of the data will be acceptable after data processing, so the number of replicate analyses required can be reasonably estimated. For challenging samples (volcanic tuffs, saline lake deposits, acicular clusters and others where petrographic examination has revealed potential sub-surface inclusions or intergrown phases), the quantity of usable data might be as low as 20% (or even non-achievable due to fine grained textures and/or persistent subsurface non-zeolite phases such as smectite clays, oxides or precursor glass), so larger numbers of analysis points need to be planned. In such circumstances, investment of time in skilful petrographic preparation becomes especially worthwhile.
2.3. Next, determine the presence of contaminants, using Fe 2 O 3 > 0.2 wt.% as a guide, and optionally, P 2 O 5 > LOD (limit of detection). For most zeolites other than faujasite, high MgO is also a suspicious indicator due to the potential presence of smectite clay minerals, ideally avoided with careful petrographic preparations. Where the R value and E% are sound, minor MgO is acceptable in the analysis. Delete the contaminant-specific components (mainly Fe 2 O 3 ) from the analyses. 2.4. Recalculate total anhydrous oxides, E% and R. 2.5. At this stage, apply the criteria of E% and R range limitations to the data, omitting analyses with E% less than -10 or greater than +10, and R values outside the mineralspecific ranges specified in table 1 of Passaglia and Sheppard (2001). 2.6. Omit further data that lie outside the range of total non-volatile oxides of between 80-95 wt.%, except for leucite (anhydrous, ideally 100%). Optionally, accept totals lower than 80 wt.%. 2.7. Next, acceptable analyses can be processed for limits of detection, determined per oxide, per mineral, per session from the average listings of the element 1-sigma data, recalculated to oxide 3-sigma values. All data components below these limits of detection are deleted, and the quality parameters of total anhydrous oxides, R and E% again recalculated. 2.8. The H 2 O component can now be estimated on the basis of 100-(total oxides)%. 2.9. Calculation of mineral formulae is undertaken according to the conventions set out in Deer et al. (2004) based on numbers of framework oxygen atoms specific to each different mineral species. Using cation proportions that were determined on the basis of 24 oxygens, and omitting all the non-formula and sub-detection components as above, re-normalize according to the mineral phase in question, i.e. 24 oxygens for chabazite, 20 for edingtonite, 96 for analcime, etc. 2.10. Calculate the sums of the framework tetrahedral components and the extraframework cation components, respectively.