Analysis of powdered reference materials and known samples with a benchtop, field portable X-ray fluorescence (pXRF) spectrometer: evaluation of performance and potential applications for exploration lithogeochemistry

Powdered international reference materials and samples with previously obtained conventional geochemical data were analysed using a benchtop portable X-ray fluorescence (pXRF) spectrometer to test the abilities of pXRF in silicate rock lithogeochemistry. Results from international reference materials illustrate that pXRF can provide very precise data for many major, minor, and trace elements, generally with RSD values of <7.5 % and many <5 %, except at very low concentrations (i.e. approaching the limit of detection). Despite good precision, accuracy is highly variable and ranges from excellent to reasonable for many major and minor elements (±15–20 % relative difference, RD, for Al2O3, SiO2, K2O, CaO, Fe2O3, TiO2, and MnO±S), base metals (±20 % for Cu, Zn), the low field strength (LFSE) and high field strength elements (HFSE) (±15 % RD for Rb, Ba, Zr; ±20 % RD for Nb). Poor accuracy was obtained for MgO, P2O5, and the transition elements (V, Cr, Ni); Sr shows variable accuracy. Comparison of pXRF results to independent samples with data from conventional analyses illustrates very poor correlation for MgO, P2O5, V, Cr, and Ni, suggesting they have poor accuracy by pXRF. Aluminum (Al2O3), SiO2, and Zn have r2 values of c. 0.6–0.7 illustrating reasonable correlation, whereas most other elements (S, K2O, CaO, TiO2, MnO, Fe2O3, Co, Cu, Pb, Rb, Sr, Ba, Zr, Nb, U, As, and Mo) have very good to excellent correlation between pXRF data and conventional analysis (i.e. r2 >0.80). In addition, many of the elements with r2 >0.8 have slopes that are close to 1 or within 20 % of ideal, indicating that pXRF is replicating the results of conventional analyses and likely within ±20 % of what can be obtained by conventional methods. Down-hole profiles of pXRF data and element ratios replicate the geometry of the profiles from conventional analyses and illustrate the ability of the pXRF to discriminate rock type, alteration, and mineralization in unknown samples. Portable XRF can provide fit-for-purpose data that is useful in discriminating lithogeochemical variations related to lithology, alteration, and mineralization. However, pXRF should be considered a preliminary screening tool for sample selection and not a substitute for conventional lithogeochemical methods (e.g. XRF, fusion ICP-ES and ICP-MS), particularly when important economic decisions are to be made using such data (e.g. NI-43-101 resource calculations). Supplementary Material: Collated data for repeat analyses of reference materials in Mining Plus (Table 1) and Soil 3 Beam (Table 2) modes. Tables 3-9 contain plots comparing results from pXRF to accepted values for various reference materials. All tables are available at at www.geolsoc.org.uk/SUP18735

between data acquisition and the utilization of results in active exploration and development. Field-portable X-ray fluorescence (pXRF) technology, however, has the potential to provide explorationists with the ability to obtain real time lithogeochemical data, thereby allowing immediate decisions to be made about rock types, alteration, mineralization, and the exploration potential of an area or project.
Despite the development of pXRF units, there is remarkably little published information on the utilization of these instruments in exploration lithogeochemistry and exploration geochemistry. Abundant research has been undertaken on the development of pXRF as an analytical tool (e.g. Potts et al. 1995;Potts et al. 1997a;Potts et al. 1997b;Hou et al. 2004;Potts & West 2008), in environmental research and contaminated soil analysis (e.g. Argyraki et al. 1997;Kalnicky & Singhvi 2001;Ramsey 2008;Haffert & Craw 2009;Kenna et al. 2011;Higueras et al. 2012), and geoarcheaology (e.g. Williams-Thorpe 2008; Liritzis & Zacharias 2011). Only recently has there been a concerted effort towards their utilization in exploration lithogeochemistry and geochemistry (e.g. Morris 2009; Gazley et al. 2011;Hall et al. 2012). Furthermore, the presence of pXRF in lithogeochemistry and mineral exploration is growing due to the increasing availability and commercialization of pXRF instruments by various providers.
Despite their availability and expanded usage, until recently there has been insufficient documentation of the operation, calibration, data quality, and independent testing of pXRF instruments (Hall et al. 2012). This manuscript presents the results of a study of whole rock lithogeochemical powders for international reference materials and previously analysed samples with rock types, matrixes, and compositional ranges commonly seen in exploration lithogeochemistry. We have utilized a single point calibration approach akin to the method of Longerich (1995) for conventional XRF. We provide estimates of precision and accuracy of the various international reference materials, for previously analysed standards, and discuss the potential application of pXRF data in exploration lithogeochemistry.

MethoDology
Data were obtained using a field portable benchtop Olympus/ Innov-X X5000 energy dispersive pXRF equipped with a Ta X-ray tube with a maximum tube voltage of 50 kV, maximum tube power of 10 W, detector area of 25 mm 2 , and <165 eV spectral resolution. The samples were analysed using both the 2-beam mining plus mode and the 3-beam soil mode. Elements determined in the mining mode were measured at 50 kV (V, Cr, Co, Ni, Cu, Zn, W, As, Pb, Bi, Zr, Mo, Ag, Cd, Sn, Sb, and Ti, Mn, and Fe) and 10 kV (Mg, Al, Si, P, S, Cl, K, Ca, Ti, Mn, Fe). Elements determined in the soil mode were measured at 50 kV (U,Sr,Zr,Th,Nb,Mo,Rh,Pd,Ag,Cd,Sn,Sb,Ba,and Ti,V,Mn,Fe,Co,Cu,Zn,Hg,As,Se,Tl,Pb,Bi,Rb,Nd),35 kV (Fe,Co,Ni,Cu,Zn,Hg,As,Se,Tl,Pb,Bi,Rb,and Ti,V,Cr,Mn,U,Sr,Zr,Th,Nb,Mo,Rh,Pd,Ag,Cd,Sn,Sb), and 15 kV (P, S, Cl, K, Ca, Ti, V, Cr, Mn, and Fe). Elements in italics above are those collected at that voltage but this is not their primary beam for data collection. Beam times were 60 seconds for all voltages in both soil and mining modes (White & Goodwin 2011;Hall et al. 2012). The limits of detection for the various elements by the instrument are generally in the ppm range, except for MgO, SiO 2 , and Al 2 O 3 , which are generally >0.5 % (e.g. Hall et al. 2012).
The materials analysed were in powdered form and were c. 200 mesh (74 µm) or finer. Having materials in powder form ensured homogeneity of the material and therefore removes the potential particle size effects associated with the pXRF results (e.g. Potts et al. 1997b;Markowicz 2008). Powdered samples were placed in 30.9 mm sample cups with 6 µm-Prolene thin films. Samples were shaken prior to analysis by pXRF to ensure continued homogenization (e.g. Hall et al. 2012).
The X5000 was calibrated using a single point calibration approach (rather than calibration lines) and utilized certain standards for certain elements, akin to approaches used by some labs for conventional XRF (e.g. Longerich 1995). Standards chosen for the single point calibration were of similar matrix and ranges of compositions as the materials being analysed as unknowns. The single point calibration approach was chosen as it increases the rapidity whereby the instrument can be calibrated, which in turn allows the instrument to be calibrated multiple times during an analytical session so as to minimize drift, and also allows much quicker data processing. The single point calibration approach requires that the samples being analysed are of similar matrix to the calibration standards and have concentration values similar to the calibration standards. The speed of calibration and data processing using this method, however, are reasonable trade-offs to provide rapid, fit-for-purpose data for real time decision-making. Nevertheless, samples with different matrixes and for which the anticipated values are much higher (or lower) in concentration than the calibration standards (i.e. order(s) of magnitude different) should utilize different reference materials to undertake single point calibrations (e.g. samples with ore-grade metal values should use an ore metal standard to get reasonable metal grades).
The calibration standards used for the calibrations in this study include: Japan Geological Survey reference material JR-1 (rhyolite), United States Geological Survey (USGS) reference materials DTS-2b (dunite) and BHVO-2 (basalt), and Canadian Certified Reference Materials Program (CCRMP) reference materials SY-3 (syenite) and PACS-1 (marine sediment). These standards were chosen as they have similar concentrations and matrixes to the reference materials and samples being analysed in the study. The list of calibration values and calibration standards are shown in Table 1. A silica blank provided by the manufacturer in a similar XRF cup with a Prolene thin film was utilized to monitor contamination.
A typical run involved the analysis of the five reference materials used as calibration standards, a blank, 20 unknown samples, and a blank. The procedure was then repeated for the next suite of samples.
Data processing of blank and unknown data was done in single batches with the calibration standards utilized for 20 unknowns and 2 blanks. Raw values obtained for all data were converted from elemental to oxide form for major and minor elements (e.g. Mg, Al, Si, P, K, Ca, Mn, Ti, and Fe); all other elements were left in elemental form. The raw data were then calibration-corrected using the single point calibration. To obtain a calibration-correction, the raw data for a given element from the calibration standards ([w] reference material, raw ) were divided by literature values ([w] literature ) to provide a correction factor (Corr i ) for a given element i: This correction factor was then applied to the raw concentration of element i in the unknown sample or blank j (i.e. [i] raw,i ) to provide a calibration corrected concentration for element i in sample j:

APPliCAtion to PowDeReD lithogeoCheMiCAl MAteRiAlS
To test the capabilities of the X5000 we have undertaken three experiments. The first experiment involves the utilization of international and internal reference materials run as unknowns and results were compared to accepted values from the literature. The second test compared powders for lithogeochemical samples from the archives of the senior author that had conventional lithogeochemical data. While these materials had elements measured by a variety of methods (e.g. ICP-ES, ICP-MS, conventional XRF), they do provide a second independent test of the potential utility of pXRF. A third experiment involved a series of contiguous samples from a single drill-core that have had independent conventional analyses. This experiment was designed to illustrate the potential utility and ability to replicate spatial relationships in an exploration/development environment. All pXRF analytical work was undertaken between October and December 2012.
Precision and accuracy results for the various reference materials are shown in Supplementary Data Tables 1 and 2, a summary of which is provided in Table 2, and representative examples of precision are shown in Figure 1. Precision was measured using the percent relative standard deviation (% RSD i ). Using the criteria of Jenner (1996), we suggest that excellent precision is 0-3 % RSD, very good precision is 3-7 % RSD, good precision is 7-10 % RSD, and greater than 10 % RSD is poor precision. Accuracy was determined using the percent relative difference (% RD) between elemental values found in the reference materials versus that in the literature. The literature values for the reference materials were obtained from the original USGS certificates, from the Govindaraju (1994) archive of values, and more recent updates from the online GeoREM archive (http://http:// georem.mpch-mainz.gwdg.de). The source of the accepted values for each standard is noted in Appendices 1 and 2.
Precision. Precision by pXRF is good to excellent for most elements, especially those with elevated concentrations ( Fig. 1; Table 2; Supplementary Data Tables 1, 2). To remove potential precision outliers due to low concentrations, in Table 2 we have summarized the precision for a given element where the element concentration is above the limit of quantification (LOQ = 3.3x the limit of detection) and 10 times the limit of detection. This allows a more robust evaluation of precision than if the entire dataset is utilized. With minor exceptions 3.79 to 11.9 -1.33 to -50.9 3.79 to 9.29 -1.33 to -50.9 S 5.44 to 6.11 -6.50 to -10.0 5.44 to 6.11 -6.50 to -9. --1 ranges shown are for average values for the reference materials that are above the limit limit of quantification (LOQ = 3.3 × limit of detection). 2 ranges shown are for average values for the reference materials that are 10 times the limit of detection. Limit of detection used for LOQ and 10× values are from Hall et al. (2012). Accepted values of concentrations for the reference materials were utilized to determine relationship of results to detection limit. *Only a few values are above LOQ and data not reliable.
precision is good to excellent for most elements with many samples having an RSD of <7.5 % (Fig. 1b). Poorer precision exists for MgO, Ni, Ba, and for elements with only a few analyses above the LOQ (e.g. Pb, U, Mo). In some cases precision decreases with concentration (e.g. TiO 2 , P 2 O 5 ; Fig. 1c). Accuracy. While precision is good across most elements, the accuracy of pXRF data for the international reference materials exhibits significant variability ( Fig. 1; Table 2; Supplementary  Data Tables 1, 2). The light elements (MgO, Al 2 O 3 , SiO 2 , P 2 O 5 , and S) exhibit very high variability in accuracy unless they are 10x the LOD, and even then the accuracy is poor for many reference materials (Fig. 1d)  conventional analyses (e.g. Longerich 1995). The transition metals (V, Cr, Co, and Ni) show highly variable accuracy, but are generally poor, as are base metals (Cu, Zn), albeit to a lesser extent. The elements Rb, Sr, Zr, Ba, and Nb provide surprisingly reasonable results and are roughly ±20 % RD. Arsenic, Mo, Pb, and U have insufficient results to elaborate on quality; however, given their low concentrations in most surface materials it is assumed that their accuracy would be poor by pXRF when at low concentrations.

Comparisons to independently analysed whole rock samples
The goal of this experiment is to compare results from conventional methods and pXRF and to evaluate whether pXRF can replicate lithogeochemical patterns, if not absolute data, for powders/rocks in real time. Conventional data from the unknown samples were obtained by total methods, either via utilization of solid source methods, including pressed pellet and fused disc conventional XRF, or total digestion methods (e.g. fusion or bomb dissolution) and subsequent analysis by inductively coupled plasma emission spectroscopy (ICP-ES) and/or inductively coupled plasma mass spectrometry (ICP-MS  (2001); a sample of subalkalic (tholeiitic) greenstone from the Yukon-Tanana terrane, Yukon (Piercey et al. 2004); within-plate, weakly alkalic VMS-related rhyolite flows to porphyritic intrusive rocks from the Yukon-Tanana terrane (Piercey et al. , 2008; and two sulphide-and metal-rich hydrothermal mudstones from the Lemarchant VMS deposit, central Newfoundland (Piercey, unpublished data). Representative examples of element behavior are shown in graphical form in Figure 2 (MgO, SiO 2 , Fe 2 O 3 ) and in Supplementary Data Tables 3-9, and linear relationships between the conventional analyses and pXRF results and their relevant coefficients of determination (r 2 ) are given in Table 3. These results corroborate many of the results from the international reference materials and illustrate extremely poor correlations for MgO, P 2 O 5 , V, Cr, and Ni, suggesting they have poor accuracy. Oxides Al 2 O 3 and SiO 2 have r 2 values between 0.60 and 0.70 and all other elements have r 2 values, when filtered for outliers, that are >0.8, implying very significant correlations. Furthermore, most of the samples with r 2 values of >0.8 have slopes that are ±20 % of 1, with the exception of U and Mo, implying that pXRF is replicating the results of conventional analyses by c. ±20 %.

Applications to chemostratigraphy
To test the potential utilization of pXRF data in a spatial context, a down-hole profile of pXRF data relative to conventional data, comprising both element concentrations and element ratios, is shown in Figures 3-6. The drill-hole utilized is from the Ming VMS deposit, Newfoundland, Canada, and was chosen because the hole had a wide range of concentrations of major and trace elements due to lithologic variation, alteration, and mineralization. Therefore, this drill-hole tests whether pXRF can replicate spatial lithogeochemical relationships in an exploration context. Samples were collected during mapping of the drill-core in 2010 and were prepared by conventional crushing and pulverizing at a commercial laboratory. These powders were then analysed by fusion ICP-ES for major elements, and by bomb dissolution and ICP-MS for trace elements (i.e. HFSE, LFSE, metals) and rare earth elements. These same powders were then used for pXRF determinations.
For major elements, the pXRF data for Al 2 O 3 and MgO show broadly similar shapes to the conventional data; however, there is significant offset in the results (Fig. 3). In contrast, SiO 2 , K 2 O, and Fe 2 O 3 are much better with most results nearly identical to the conventional data (Fig. 3). Elements that are indicative of mineralization (e.g. base metals Cu, Zn, and Pb), or are pathfinder elements (As, Sb), have variable behavior (Fig. 4). The base metals (Cu, Zn, and Pb) and As replicate the results from conventional data but with some offset, whereas Sb only approximates the conventional data (Fig. 4). Elements typically used for chemostratigraphy (e.g. Barrett & MacLean 1999;Piercey 2009) show fairly good coincidence with conventional data, except for values that are approaching the LOD/LOQ of pXRF (e.g. Nb; Fig. 5). Similarly, alteration-sensitive mobile elements and element ratios (e.g. Barrett & MacLean 1999;Piercey 2009) yield similar patterns and considerable overlap between pXRF and conventional data, except for those that are close or below the LOD/LOQ (e.g. some Rb and Ba values) (Fig. 6).

DiSCuSSion AnD SuMMARy
The single point calibration method provided in this paper illustrates that calibration using a minimal number of calibration standards results in reasonably good precision and accuracy. While it is acknowledged that multiple calibration points per element (i.e. calibration lines) would likely extend the dynamic range of precision and accuracy, the results we have obtained on international reference materials are very similar to the more extensive study of Hall et al. (2012). Replicate analyses of international reference materials using the X5000 are excellent to good for most major to minor elements (K 2 O, CaO, TiO 2 , MnO) including many light elements (Al 2 O 3 , SiO 2 , P 2 O 5 ). In contrast, the major element MgO has poor precision, and there are insufficient data to evaluate the precision of S and Cl. Similarly, the X5000 can produce precise data for trace elements, including transition metals (V, Cr, Co), base metals (Cu, Zn), LFSE (Rb, Sr), and HFSE (Zr, Nb), except at lower concentrations (i.e. approaching LOD and LOQ), with most having RSD values below 7.5 % and many with RSD values below 5 %. The exceptions include Ni, which has limited data and low precision, and Ba, which is also less precise.
While the precision of the X5000 is good for most elements, the accuracy is variable. Most reference materials have accuracy of ±15-20 % RD for the major and minor elements Al 2 O 3 , SiO 2 , K 2 O, CaO, Fe 2 O 3 , TiO 2 , and MnO, except where the elements are at low concentrations. Sulphur shows promise as the two reference materials with S >1.5 wt% (>15000 ppm) have an accuracy that is <10% RD. In contrast, MgO and P 2 O 5 have poor accuracy. The transition metals have very poor accuracy, whereas base metals Cu and Zn have 20 % RD or better. The poor accuracy for the transition elements is likely due to X-ray peak overlaps that are not resolved by the X5000 software (e.g. Fe Kα and Kβ peaks overlapping with Co Kα peak) (Thompson et al. 2009;Hall et al. 2012). The lower accuracy for the base metals is possibly due to matrix effects, as most base metals in the reference materials are likely associated with sulphides, whereas the calibration standard for the base metals is a silicate matrix (i.e. basaltic). Using a calibration standard with a sulphide matrix   Fig. 3. Down-hole profile for key major elements (in %) used in the discrimination of lithology, alteration, and mineralization. MP suffix at the end of the element implies analysis by mining mode. Elements without suffixes were data obtained by conventional methods (fusion ICP-ES and ICP-MS).

Fig. 4.
Down-hole profile for key trace elements (in ppm) used in delineating mineralization, including metals (Cu, Zn, Pb) and metalloids (As, Sb). S3B suffix at the end of the element implies analysis by soil mode. Elements without suffixes were data obtained by conventional methods (fusion ICP-ES and ICP-MS).

Fig. 5.
Down-hole profile for key major (in %) and trace elements (in ppm) and element ratios used in lithologic discrimination and chemostratigraphy (e.g. Winchester & Floyd 1977;Barrett & MacLean 1999). Suffixes S3B and pXRF reflect data obtained by the soil mode and ratios of pXRF data, respectively. Elements and element ratios without suffixes were data obtained by conventional methods (fusion ICP-ES and ICP-MS). may improve accuracy and is the focus of ongoing research by the Memorial University group. The LFSE and HFSE Rb, Ba, and Zr have accuracy in the range of ±15 % RD, whereas Nb is 20 % RD or better (with minor exceptions) and Sr is variable. The variability of Sr accuracy contrasts with other research as it is one of the most accurate elements during pXRF testing by Morris (2009) and Hall et al. (2012). Further, Sr shows very strong correlations with slopes close to 1, when compared to samples with independent conventional geochemical analyses for Sr.
The results for international reference materials are also replicated in comparisons of pXRF results on samples with existing analyses by conventional geochemical methods. There are extremely poor correlations for MgO, P 2 O 5 , V, Cr, and Ni, suggesting they have poor accuracy. In contrast, there are strong correlations between pXRF data and conventional analysis (i.e. r 2 >0.80) for S, K 2 O, CaO, TiO 2 , MnO, Fe 2 O 3 , Co, Cu, Pb, Rb, Sr, Ba, Zr, Nb, U, As, and Mo. Despite some scatter, there are also weaker correlations for Al 2 O 3 , SiO 2 , and Zn (Table 3; Supplementary Data Tables 3-9). In addition, many of these elements have slopes that are close to 1 or within 20 % of ideal, indicating that pXRF is replicating the results of conventional analyses for the aforementioned elements. Furthermore, the down-hole profiles of various elements that are critical for lithogeochemical identification of rock type, mineralization, and alteration are clearly replicated and there is overlap between conventional and pXRF data.
The results of this study illustrate that pXRF can provide reasonable quality (i.e. fit-for-purpose) data for many elements. Furthermore, it can replicate the results of conventional methods and can be utilized in exploration lithogeochemistry. It is clear, however, that pXRF cannot provide data that are a substitute for more robust conventional methods like traditional fused disc or pressed pellet XRF and fusion ICP-ES and ICP-MS (e.g. Hall et al. 2012), particularly when economic decisions are to be made (e.g. NI-43-101 resource calculations). Nevertheless, it has a potentially very useful future in the screening samples for which to obtain conventional data and in preliminary chemostratigraphic and alteration studies.
The Olympus-Innov-X 5000 at Memorial University was purchased through funds from the GeoExplore Program from the Research and Development Corporation of Newfoundland and Labrador (RDCNL). This research was also supported by an NSERC Discovery Grant, and the NSERC-Altius Industrial Research Chair in Mineral Deposits supported by NSERC, Altius Resources Ltd., and RDCNL. Olympus-Innov-X are thanked for their support and technical assistance during the early stages of pXRF development. Jim Ryan of the Geological Survey of Canada provided the Multinational Andean Project internal reference materials. Jon Oliver and Dean Courage are thanked for assistance with sample preparation and analysis. Discussions and correspondence with Gwendy Hall, George Jenner, Henry Longerich, Trevor MacHattie, and Jan Peter throughout this project were greatly appreciated. Gwendy Hall and an anonymous reviewer are thanked for their very thorough reviews and helpful suggestions.

ReFeRenCeS
ArgyrAki, A., H. rAmsey, M. & J. Potts, P. 1997 Fig. 6. Down-hole profile for key major (in %) and trace elements (in ppm) and element ratios used in documenting alteration (e.g. Barrett & MacLean 1999;Piercey 2009). Suffixes S3B and pXRF reflect data obtained by the soil mode and ratios of pXRF data, respectively. Elements and element ratios without suffixes were data obtained by conventional methods (fusion ICP-ES and ICP-MS).